release : v1.2.0

* ggml : add "scratch" buffer support

* ggml : support for scratch ring-buffer

* ggml : bug fix in ggml_repeat()

* ggml : error on scratch buffer overflow

* whisper : use scratch buffers during inference (base model only)

* whisper : update memory usage for all models

* whisper : fix encoder memory usage

* whisper : use whisper_context functions instead of macros

* whisper : fix FF + remove it from README

* ggml : reuse ggml_new_i32

* ggml : refactor the scratch buffer storage

* whisper : reorder scratch buffers in the decoder

* main : add option to disable temp fallback

* Update README.md

.github/workflows/bindings.yml

@@ -0,0 +1,22 @@
+name: Bindings Tests
+on:
+  push:
+    paths:
+      - bindings/go/**
+      - whisper.h
+  pull_request:
+    paths:
+      - bindings/go/**
+      - whisper.h
+
+jobs:
+  ubuntu-latest:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/setup-go@v3
+        with:
+          go-version: '^1.19'
+      - uses: actions/checkout@v1
+      - run: |
+          cd bindings/go
+          make test

.github/workflows/build.yml

@@ -1,237 +1,267 @@
 name: CI
-on: [push]
+on: [push, pull_request]
 
 jobs:
-    ubuntu-latest:
-        runs-on: ubuntu-latest
+  ubuntu-latest:
+    runs-on: ubuntu-latest
 
-        steps:
-            - name: Clone
-              uses: actions/checkout@v1
+    steps:
+      - name: Clone
+        uses: actions/checkout@v1
 
-            - name: Dependencies
-              run: |
-                  sudo apt-get update
-                  sudo apt-get install build-essential
-                  sudo apt-get install libsdl2-dev
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential
+          sudo apt-get install libsdl2-dev
 
-            - name: Build
-              run: |
-                make
-                make stream
+      - name: Build
+        run: |
+          make
+          make stream
 
-    macOS-latest:
-        runs-on: macOS-latest
+  macOS-latest:
+    runs-on: macOS-latest
 
-        steps:
-            - name: Clone
-              uses: actions/checkout@v1
+    steps:
+      - name: Clone
+        uses: actions/checkout@v1
 
-            - name: Dependencies
-              run: |
-                  brew update
-                  brew install sdl2
+      - name: Dependencies
+        run: |
+          brew update
+          brew install sdl2
 
-            - name: Build
-              run: |
-                make
-                make stream
+      - name: Build
+        run: |
+          make
+          make stream
 
-    ubuntu-latest-gcc:
-        runs-on: ubuntu-latest
+  ubuntu-latest-gcc:
+    runs-on: ubuntu-latest
 
-        strategy:
-            matrix:
-                build: [Debug, Release]
+    strategy:
+      matrix:
+        build: [Debug, Release]
 
-        steps:
-            - name: Clone
-              uses: actions/checkout@v1
+    steps:
+      - name: Clone
+        uses: actions/checkout@v1
 
-            - name: Dependencies
-              run: |
-                  sudo apt-get update
-                  sudo apt-get install build-essential
-                  sudo apt-get install cmake
-                  sudo apt-get install libsdl2-dev
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential
+          sudo apt-get install cmake
+          sudo apt-get install libsdl2-dev
 
-            - name: Configure
-              run: cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }}
+      - name: Configure
+        run: cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }}
 
-            - name: Build
-              run: |
-                make
-                ctest -L gh --output-on-failure
+      - name: Build
+        run: |
+          make
+          ctest -L gh --output-on-failure
 
-    ubuntu-latest-clang:
-        runs-on: ubuntu-latest
+  ubuntu-latest-clang:
+    runs-on: ubuntu-latest
 
-        strategy:
-            matrix:
-                build: [Debug, Release]
+    strategy:
+      matrix:
+        build: [Debug, Release]
 
-        steps:
-            - name: Clone
-              uses: actions/checkout@v1
+    steps:
+      - name: Clone
+        uses: actions/checkout@v1
 
-            - name: Dependencies
-              run: |
-                  sudo apt-get update
-                  sudo apt-get install build-essential
-                  sudo apt-get install cmake
-                  sudo apt-get install libsdl2-dev
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential
+          sudo apt-get install cmake
+          sudo apt-get install libsdl2-dev
 
-            - name: Configure
-              run: cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang
+      - name: Configure
+        run: cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang
 
-            - name: Build
-              run: |
-                make
-                ctest -L gh --output-on-failure
+      - name: Build
+        run: |
+          make
+          ctest -L gh --output-on-failure
 
-    ubuntu-latest-gcc-sanitized:
-        runs-on: ubuntu-latest
+  ubuntu-latest-gcc-sanitized:
+    runs-on: ubuntu-latest
 
-        strategy:
-            matrix:
-                sanitizer: [ADDRESS, THREAD, UNDEFINED]
+    strategy:
+      matrix:
+        sanitizer: [ADDRESS, THREAD, UNDEFINED]
 
-        steps:
-            - name: Clone
-              uses: actions/checkout@v1
+    steps:
+      - name: Clone
+        uses: actions/checkout@v1
 
-            - name: Dependencies
-              run: |
-                  sudo apt-get update
-                  sudo apt-get install build-essential
-                  sudo apt-get install cmake
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential
+          sudo apt-get install cmake
 
-            - name: Configure
-              run: cmake . -DCMAKE_BUILD_TYPE=Debug -DWHISPER_SANITIZE_${{ matrix.sanitizer }}=ON
+      - name: Configure
+        run: cmake . -DCMAKE_BUILD_TYPE=Debug -DWHISPER_SANITIZE_${{ matrix.sanitizer }}=ON
 
-            - name: Build
-              run: |
-                make
-                ctest -L gh --output-on-failure
+      - name: Build
+        run: |
+          make
+          ctest -L gh --output-on-failure
 
-    windows:
-        runs-on: windows-latest
+  windows:
+    runs-on: windows-latest
 
-        strategy:
-            matrix:
-                build: [Release]
-                arch: [Win32, x64]
-                sdl2: [ON]
-                include:
-                  - arch: Win32
-                    s2arc: x86
-                  - arch: x64
-                    s2arc: x64
-                  - sdl2: ON
-                    s2ver: 2.26.0
+    strategy:
+      matrix:
+        build: [Release]
+        arch: [Win32, x64]
+        sdl2: [ON]
+        include:
+          - arch: Win32
+            s2arc: x86
+          - arch: x64
+            s2arc: x64
+          - sdl2: ON
+            s2ver: 2.26.0
 
-        steps:
-            - name: Clone
-              uses: actions/checkout@v1
+    steps:
+      - name: Clone
+        uses: actions/checkout@v1
 
-            - name: Add msbuild to PATH
-              uses: microsoft/setup-msbuild@v1
+      - name: Add msbuild to PATH
+        uses: microsoft/setup-msbuild@v1
 
-            - name: Fetch SDL2 and set SDL2_DIR
-              if: matrix.sdl2 == 'ON'
-              run: |
-                C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
-                7z x sdl2.zip
-                echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
+      - name: Fetch SDL2 and set SDL2_DIR
+        if: matrix.sdl2 == 'ON'
+        run: |
+          C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
+          7z x sdl2.zip
+          echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
 
-            - name: Configure
-              run: >
-                cmake -S . -B ./build -A ${{ matrix.arch }}
-                -DCMAKE_BUILD_TYPE=${{ matrix.build }}
-                -DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
+      - name: Configure
+        run: >
+          cmake -S . -B ./build -A ${{ matrix.arch }}
+          -DCMAKE_BUILD_TYPE=${{ matrix.build }}
+          -DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
 
-            - name: Build
-              run: |
-                cd ./build
-                msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
+      - name: Build
+        run: |
+          cd ./build
+          msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
 
-            - name: Copy SDL2.dll
-              if: matrix.sdl2 == 'ON'
-              run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
+      - name: Copy SDL2.dll
+        if: matrix.sdl2 == 'ON'
+        run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
 
-            - name: Upload binaries
-              if: matrix.sdl2 == 'ON'
-              uses: actions/upload-artifact@v1
-              with:
-                name: whisper-bin-${{ matrix.arch }}
-                path: build/bin/${{ matrix.build }}
+      - name: Upload binaries
+        if: matrix.sdl2 == 'ON'
+        uses: actions/upload-artifact@v1
+        with:
+          name: whisper-bin-${{ matrix.arch }}
+          path: build/bin/${{ matrix.build }}
 
-    windows-blas:
-        runs-on: windows-latest
+  windows-blas:
+    runs-on: windows-latest
 
-        strategy:
-            matrix:
-                build: [Release]
-                arch: [Win32, x64]
-                blas: [ON]
-                sdl2: [ON]
-                include:
-                  - arch: Win32
-                    obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x86.zip
-                    s2arc: x86
-                  - arch: x64
-                    obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x64.zip
-                    s2arc: x64
-                  - sdl2: ON
-                    s2ver: 2.26.0
+    strategy:
+      matrix:
+        build: [Release]
+        arch: [Win32, x64]
+        blas: [ON]
+        sdl2: [ON]
+        include:
+          - arch: Win32
+            obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x86.zip
+            s2arc: x86
+          - arch: x64
+            obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x64.zip
+            s2arc: x64
+          - sdl2: ON
+            s2ver: 2.26.0
 
-        steps:
-            - name: Clone
-              uses: actions/checkout@v1
+    steps:
+      - name: Clone
+        uses: actions/checkout@v1
 
-            - name: Add msbuild to PATH
-              uses: microsoft/setup-msbuild@v1
+      - name: Add msbuild to PATH
+        uses: microsoft/setup-msbuild@v1
 
-            - name: Fetch OpenBLAS
-              if: matrix.blas == 'ON'
-              run: |
-                C:/msys64/usr/bin/wget.exe -qO blas.zip ${{ matrix.obzip }}
-                7z x blas.zip -oblas -y
-                copy blas/include/cblas.h .
-                copy blas/include/openblas_config.h .
-                echo "blasdir=$env:GITHUB_WORKSPACE/blas" >> $env:GITHUB_ENV
+      - name: Fetch OpenBLAS
+        if: matrix.blas == 'ON'
+        run: |
+          C:/msys64/usr/bin/wget.exe -qO blas.zip ${{ matrix.obzip }}
+          7z x blas.zip -oblas -y
+          copy blas/include/cblas.h .
+          copy blas/include/openblas_config.h .
+          echo "blasdir=$env:GITHUB_WORKSPACE/blas" >> $env:GITHUB_ENV
 
-            - name: Fetch SDL2 and set SDL2_DIR
-              if: matrix.sdl2 == 'ON'
-              run: |
-                C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
-                7z x sdl2.zip
-                echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
+      - name: Fetch SDL2 and set SDL2_DIR
+        if: matrix.sdl2 == 'ON'
+        run: |
+          C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
+          7z x sdl2.zip
+          echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
 
-            - name: Configure
-              run: >
-                cmake -S . -B ./build -A ${{ matrix.arch }}
-                -DCMAKE_BUILD_TYPE=${{ matrix.build }}
-                -DWHISPER_SUPPORT_OPENBLAS=${{ matrix.blas }}
-                -DCMAKE_LIBRARY_PATH="$env:blasdir/lib"
-                -DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
+      - name: Configure
+        run: >
+          cmake -S . -B ./build -A ${{ matrix.arch }}
+          -DCMAKE_BUILD_TYPE=${{ matrix.build }}
+          -DWHISPER_SUPPORT_OPENBLAS=${{ matrix.blas }}
+          -DCMAKE_LIBRARY_PATH="$env:blasdir/lib"
+          -DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
 
-            - name: Build
-              run: |
-                cd ./build
-                msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
+      - name: Build
+        run: |
+          cd ./build
+          msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
 
-            - name: Copy libopenblas.dll
-              if: matrix.blas == 'ON'
-              run: copy "$env:blasdir/bin/libopenblas.dll" build/bin/${{ matrix.build }}
+      - name: Copy libopenblas.dll
+        if: matrix.blas == 'ON'
+        run: copy "$env:blasdir/bin/libopenblas.dll" build/bin/${{ matrix.build }}
 
-            - name: Copy SDL2.dll
-              if: matrix.sdl2 == 'ON'
-              run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
+      - name: Copy SDL2.dll
+        if: matrix.sdl2 == 'ON'
+        run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
 
-            - name: Upload binaries
-              if: matrix.blas == 'ON' && matrix.sdl2 == 'ON'
-              uses: actions/upload-artifact@v1
-              with:
-                name: whisper-blas-bin-${{ matrix.arch }}
-                path: build/bin/${{ matrix.build }}
+      - name: Upload binaries
+        if: matrix.blas == 'ON' && matrix.sdl2 == 'ON'
+        uses: actions/upload-artifact@v1
+        with:
+          name: whisper-blas-bin-${{ matrix.arch }}
+          path: build/bin/${{ matrix.build }}
+
+  emscripten:
+    runs-on: ubuntu-latest
+
+    strategy:
+      matrix:
+        build: [Release]
+
+    steps:
+      - name: Clone
+        uses: actions/checkout@v1
+
+      - name: Dependencies
+        run: |
+          wget -q https://github.com/emscripten-core/emsdk/archive/master.tar.gz
+          tar -xvf master.tar.gz
+          emsdk-master/emsdk update
+          emsdk-master/emsdk install latest
+          emsdk-master/emsdk activate latest
+
+      - name: Configure
+        run: echo "tmp"
+
+      - name: Build
+        run: |
+          pushd emsdk-master
+          source ./emsdk_env.sh
+          popd
+          emcmake cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }}
+          make

.gitignore

@@ -1,4 +1,5 @@
 *.o
+*.a
 .cache/
 .vs/
 .vscode/
@@ -8,6 +9,7 @@ build/
 build-em/
 build-debug/
 build-release/
+build-static/
 build-sanitize-addr/
 build-sanitize-thread/
 
@@ -17,7 +19,9 @@ build-sanitize-thread/
 /talk
 /bench
 
+arm_neon.h
 sync.sh
+libwhisper.a
 libwhisper.so
 compile_commands.json
 

CMakeLists.txt

@@ -1,15 +1,16 @@
 cmake_minimum_required (VERSION 3.0)
 
-project(whisper.cpp VERSION 1.0.4)
+project(whisper.cpp VERSION 1.2.0)
+
+# Add path to modules
+list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
 
-set(CMAKE_EXPORT_COMPILE_COMMANDS "on")
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
-set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/lib")
 
 if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
     set(WHISPER_STANDALONE ON)
-    include(cmake/GitVars.cmake)
-    include(cmake/BuildTypes.cmake)
+    include(GitVars)
+    include(BuildTypes)
 
     # configure project version
     if (EXISTS "${CMAKE_SOURCE_DIR}/bindings/ios/Makefile-tmpl")
@@ -52,6 +53,7 @@ if (APPLE)
     option(WHISPER_NO_ACCELERATE       "whisper: disable Accelerate framework" OFF)
     option(WHISPER_NO_AVX              "whisper: disable AVX" OFF)
     option(WHISPER_NO_AVX2             "whisper: disable AVX2" OFF)
+    option(WHISPER_NO_FMA              "whisper: disable FMA" OFF)
 else()
     option(WHISPER_SUPPORT_OPENBLAS    "whisper: support for OpenBLAS" OFF)
 endif()
@@ -82,9 +84,6 @@ endif()
 
 # dependencies
 
-set(CMAKE_C_STANDARD   11)
-set(CMAKE_CXX_STANDARD 11)
-
 find_package(Threads REQUIRED)
 
 # on APPLE - include Accelerate framework
@@ -131,6 +130,13 @@ if (WHISPER_ALL_WARNINGS)
             -Wcast-qual                     \
             -Wstrict-prototypes             \
             -Wpointer-arith                 \
+            -Wno-unused-function            \
+        ")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} \
+            -Wall                           \
+            -Wextra                         \
+            -Wpedantic                      \
+            -Wcast-qual                     \
         ")
     else()
         # todo : msvc
@@ -151,6 +157,7 @@ else()
     if (MSVC)
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX2")
         set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /arch:AVX2")
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /arch:AVX2")
     else()
         if (EMSCRIPTEN)
             set(CMAKE_C_FLAGS   "${CMAKE_C_FLAGS}   -pthread")
@@ -162,7 +169,10 @@ else()
             if(NOT WHISPER_NO_AVX2)
                 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx2")
             endif()
-            set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mfma -mf16c")
+            if(NOT WHISPER_NO_FMA)
+                set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mfma")
+            endif()
+            set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mf16c")
         endif()
     endif()
 endif()
@@ -178,10 +188,14 @@ endif()
 set(TARGET whisper)
 
 add_library(${TARGET}
+    ggml.h
     ggml.c
+    whisper.h
     whisper.cpp
     )
 
+include(DefaultTargetOptions)
+
 target_include_directories(${TARGET} PUBLIC
     .
     )
@@ -215,6 +229,7 @@ target_compile_definitions(${TARGET} PUBLIC
 install(TARGETS ${TARGET}
     LIBRARY DESTINATION lib
     ARCHIVE DESTINATION lib/static
+    RUNTIME DESTINATION bin
     )
 
 #

Makefile

@@ -10,6 +10,9 @@ ifndef UNAME_M
 UNAME_M := $(shell uname -m)
 endif
 
+CCV := $(shell $(CC) --version | head -n 1)
+CXXV := $(shell $(CXX) --version | head -n 1)
+
 # Mac OS + Arm can report x86_64
 # ref: https://github.com/ggerganov/whisper.cpp/issues/66#issuecomment-1282546789
 ifeq ($(UNAME_S),Darwin)
@@ -53,10 +56,13 @@ endif
 # Architecture specific
 # TODO: probably these flags need to be tweaked on some architectures
 #       feel free to update the Makefile for your architecture and send a pull request or issue
-ifeq ($(UNAME_M),x86_64)
+ifeq ($(UNAME_M),$(filter $(UNAME_M),x86_64 i686))
 	ifeq ($(UNAME_S),Darwin)
-		CFLAGS += -mfma -mf16c
+		CFLAGS += -mf16c
 		AVX1_M := $(shell sysctl machdep.cpu.features)
+		ifneq (,$(findstring FMA,$(AVX1_M)))
+			CFLAGS += -mfma
+		endif
 		ifneq (,$(findstring AVX1.0,$(AVX1_M)))
 			CFLAGS += -mavx
 		endif
@@ -81,6 +87,10 @@ ifeq ($(UNAME_M),x86_64)
 		ifneq (,$(findstring f16c,$(F16C_M)))
 			CFLAGS += -mf16c
 		endif
+		SSE3_M := $(shell grep "sse3 " /proc/cpuinfo)
+		ifneq (,$(findstring sse3,$(SSE3_M)))
+			CFLAGS += -msse3
+		endif
 	else ifeq ($(UNAME_S),Haiku)
 		AVX1_M := $(shell sysinfo -cpu | grep "AVX ")
 		ifneq (,$(findstring avx,$(AVX1_M)))
@@ -105,6 +115,16 @@ endif
 ifeq ($(UNAME_M),amd64)
 	CFLAGS += -mavx -mavx2 -mfma -mf16c
 endif
+ifneq ($(filter ppc64%,$(UNAME_M)),)
+	POWER9_M := $(shell grep "POWER9" /proc/cpuinfo)
+	ifneq (,$(findstring POWER9,$(POWER9_M)))
+		CFLAGS += -mpower9-vector
+	endif
+	# Require c++23's std::byteswap for big-endian support.
+	ifeq ($(UNAME_M),ppc64)
+		CXXFLAGS += -std=c++23 -DGGML_BIG_ENDIAN
+	endif
+endif
 ifndef WHISPER_NO_ACCELERATE
 	# Mac M1 - include Accelerate framework
 	ifeq ($(UNAME_S),Darwin)
@@ -117,8 +137,8 @@ ifdef WHISPER_OPENBLAS
 	LDFLAGS += -lopenblas
 endif
 ifdef WHISPER_GPROF
-	CFLAGS  += -pg
-	CXXFLAGS  += -pg
+	CFLAGS   += -pg
+	CXXFLAGS += -pg
 endif
 ifneq ($(filter aarch64%,$(UNAME_M)),)
 endif
@@ -135,6 +155,21 @@ ifneq ($(filter armv8%,$(UNAME_M)),)
 	CFLAGS += -mfp16-format=ieee -mno-unaligned-access
 endif
 
+#
+# Print build information
+#
+
+$(info I whisper.cpp build info: )
+$(info I UNAME_S:  $(UNAME_S))
+$(info I UNAME_P:  $(UNAME_P))
+$(info I UNAME_M:  $(UNAME_M))
+$(info I CFLAGS:   $(CFLAGS))
+$(info I CXXFLAGS: $(CXXFLAGS))
+$(info I LDFLAGS:  $(LDFLAGS))
+$(info I CC:       $(CCV))
+$(info I CXX:      $(CXXV))
+$(info )
+
 default: main
 
 #

README.md

@@ -4,15 +4,16 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 [![npm](https://img.shields.io/npm/v/whisper.cpp.svg)](https://www.npmjs.com/package/whisper.cpp/)
 
-[Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
+Stable: [v1.2.0](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.2.0) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
 
 High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisper) automatic speech recognition (ASR) model:
 
 - Plain C/C++ implementation without dependencies
 - Apple silicon first-class citizen - optimized via Arm Neon and Accelerate framework
 - AVX intrinsics support for x86 architectures
+- VSX intrinsics support for POWER architectures
 - Mixed F16 / F32 precision
-- Low memory usage (Flash Attention + Flash Forward)
+- Low memory usage (Flash Attention)
 - Zero memory allocations at runtime
 - Runs on the CPU
 - [C-style API](https://github.com/ggerganov/whisper.cpp/blob/master/whisper.h)
@@ -70,7 +71,7 @@ Now build the [main](examples/main) example and transcribe an audio file like th
 make
 
 # transcribe an audio file
-./main -f input.wav
+./main -f samples/jfk.wav
 ```
 
 ---
@@ -88,27 +89,38 @@ c++ -I. -I./examples -O3 -std=c++11 -pthread examples/main/main.cpp whisper.o gg
 usage: ./main [options] file0.wav file1.wav ...
 
 options:
-  -h,       --help          [default] show this help message and exit
-  -t N,     --threads N     [4      ] number of threads to use during computation
-  -p N,     --processors N  [1      ] number of processors to use during computation
-  -ot N,    --offset-t N    [0      ] time offset in milliseconds
-  -on N,    --offset-n N    [0      ] segment index offset
-  -d  N,    --duration N    [0      ] duration of audio to process in milliseconds
-  -mc N,    --max-context N [-1     ] maximum number of text context tokens to store
-  -ml N,    --max-len N     [0      ] maximum segment length in characters
-  -wt N,    --word-thold N  [0.01   ] word timestamp probability threshold
-  -su,      --speed-up      [false  ] speed up audio by x2 (reduced accuracy)
-  -tr,      --translate     [false  ] translate from source language to english
-  -otxt,    --output-txt    [false  ] output result in a text file
-  -ovtt,    --output-vtt    [false  ] output result in a vtt file
-  -osrt,    --output-srt    [false  ] output result in a srt file
-  -owts,    --output-words  [false  ] output script for generating karaoke video
-  -ps,      --print-special [false  ] print special tokens
-  -pc,      --print-colors  [false  ] print colors
-  -nt,      --no-timestamps [true   ] do not print timestamps
-  -l LANG,  --language LANG [en     ] spoken language
-  -m FNAME, --model FNAME   [models/ggml-base.en.bin] model path
-  -f FNAME, --file FNAME    [       ] input WAV file path
+  -h,        --help              [default] show this help message and exit
+  -t N,      --threads N         [4      ] number of threads to use during computation
+  -p N,      --processors N      [1      ] number of processors to use during computation
+  -ot N,     --offset-t N        [0      ] time offset in milliseconds
+  -on N,     --offset-n N        [0      ] segment index offset
+  -d  N,     --duration N        [0      ] duration of audio to process in milliseconds
+  -mc N,     --max-context N     [-1     ] maximum number of text context tokens to store
+  -ml N,     --max-len N         [0      ] maximum segment length in characters
+  -bo N,     --best-of N         [5      ] number of best candidates to keep
+  -bs N,     --beam-size N       [-1     ] beam size for beam search
+  -wt N,     --word-thold N      [0.01   ] word timestamp probability threshold
+  -et N,     --entropy-thold N   [2.40   ] entropy threshold for decoder fail
+  -lpt N,    --logprob-thold N   [-1.00  ] log probability threshold for decoder fail
+  -su,       --speed-up          [false  ] speed up audio by x2 (reduced accuracy)
+  -tr,       --translate         [false  ] translate from source language to english
+  -di,       --diarize           [false  ] stereo audio diarization
+  -nf,       --no-fallback       [false  ] do not use temperature fallback while decoding
+  -otxt,     --output-txt        [false  ] output result in a text file
+  -ovtt,     --output-vtt        [false  ] output result in a vtt file
+  -osrt,     --output-srt        [false  ] output result in a srt file
+  -owts,     --output-words      [false  ] output script for generating karaoke video
+  -ocsv,     --output-csv        [false  ] output result in a CSV file
+  -of FNAME, --output-file FNAME [       ] output file path (without file extension)
+  -ps,       --print-special     [false  ] print special tokens
+  -pc,       --print-colors      [false  ] print colors
+  -pp,       --print-progress    [false  ] print progress
+  -nt,       --no-timestamps     [true   ] do not print timestamps
+  -l LANG,   --language LANG     [en     ] spoken language ('auto' for auto-detect)
+             --prompt PROMPT     [       ] initial prompt
+  -m FNAME,  --model FNAME       [models/ggml-base.en.bin] model path
+  -f FNAME,  --file FNAME        [       ] input WAV file path
+
 
 bash ./models/download-ggml-model.sh base.en
 Downloading ggml model base.en ...
@@ -127,7 +139,8 @@ Running base.en on all samples in ./samples ...
 [+] Running base.en on samples/jfk.wav ... (run 'ffplay samples/jfk.wav' to listen)
 ----------------------------------------------
 
-whisper_model_load: loading model from 'models/ggml-base.en.bin'
+whisper_init_from_file: loading model from 'models/ggml-base.en.bin'
+whisper_model_load: loading model
 whisper_model_load: n_vocab       = 51864
 whisper_model_load: n_audio_ctx   = 1500
 whisper_model_load: n_audio_state = 512
@@ -140,13 +153,14 @@ whisper_model_load: n_text_layer  = 6
 whisper_model_load: n_mels        = 80
 whisper_model_load: f16           = 1
 whisper_model_load: type          = 2
+whisper_model_load: mem required  =  215.00 MB (+    6.00 MB per decoder)
+whisper_model_load: kv self size  =    5.25 MB
+whisper_model_load: kv cross size =   17.58 MB
 whisper_model_load: adding 1607 extra tokens
-whisper_model_load: mem_required  =  506.00 MB
-whisper_model_load: ggml ctx size =  140.60 MB
-whisper_model_load: memory size   =   22.83 MB
+whisper_model_load: model ctx     =  140.60 MB
 whisper_model_load: model size    =  140.54 MB
 
-system_info: n_threads = 4 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |
+system_info: n_threads = 4 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 |
 
 main: processing 'samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
 
@@ -154,12 +168,13 @@ main: processing 'samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 proc
 [00:00:00.000 --> 00:00:11.000]   And so my fellow Americans, ask not what your country can do for you, ask what you can do for your country.
 
 
-whisper_print_timings:     load time =   105.91 ms
-whisper_print_timings:      mel time =    24.62 ms
-whisper_print_timings:   sample time =     3.63 ms
-whisper_print_timings:   encode time =   324.71 ms / 54.12 ms per layer
-whisper_print_timings:   decode time =    83.58 ms / 13.93 ms per layer
-whisper_print_timings:    total time =   542.81 ms
+whisper_print_timings:     fallbacks =   0 p /   0 h
+whisper_print_timings:     load time =   113.81 ms
+whisper_print_timings:      mel time =    15.40 ms
+whisper_print_timings:   sample time =    11.58 ms /    27 runs (    0.43 ms per run)
+whisper_print_timings:   encode time =   266.60 ms /     1 runs (  266.60 ms per run)
+whisper_print_timings:   decode time =    66.11 ms /    27 runs (    2.45 ms per run)
+whisper_print_timings:    total time =   476.31 ms
 ```
 
 The command downloads the `base.en` model converted to custom `ggml` format and runs the inference on all `.wav` samples in the folder `samples`.
@@ -202,26 +217,16 @@ make large
 
 | Model  | Disk   | Mem     | SHA                                        |
 | ---    | ---    | ---     | ---                                        |
-| tiny   |  75 MB | ~390 MB | `bd577a113a864445d4c299885e0cb97d4ba92b5f` |
-| base   | 142 MB | ~500 MB | `465707469ff3a37a2b9b8d8f89f2f99de7299dac` |
-| small  | 466 MB | ~1.0 GB | `55356645c2b361a969dfd0ef2c5a50d530afd8d5` |
-| medium | 1.5 GB | ~2.6 GB | `fd9727b6e1217c2f614f9b698455c4ffd82463b4` |
-| large  | 2.9 GB | ~4.7 GB | `0f4c8e34f21cf1a914c59d8b3ce882345ad349d6` |
+| tiny   |  75 MB | ~125 MB | `bd577a113a864445d4c299885e0cb97d4ba92b5f` |
+| base   | 142 MB | ~210 MB | `465707469ff3a37a2b9b8d8f89f2f99de7299dac` |
+| small  | 466 MB | ~600 MB | `55356645c2b361a969dfd0ef2c5a50d530afd8d5` |
+| medium | 1.5 GB | ~1.7 GB | `fd9727b6e1217c2f614f9b698455c4ffd82463b4` |
+| large  | 2.9 GB | ~3.3 GB | `0f4c8e34f21cf1a914c59d8b3ce882345ad349d6` |
 
 ## Limitations
 
 - Inference only
-- No GPU support
-- Very basic greedy sampling scheme - always pick up the token with highest probability.
-  This should be similar to the [GreedyDecoder](https://github.com/openai/whisper/blob/main/whisper/decoding.py#L249-L274)
-  from the original python implementation, so in order to make a fair comparison between the 2 implementations, make sure
-  to run the python code with the following parameters:
-
-  ```
-  whisper --best_of None --beam_size None ...
-  ```
-
-  In the future, `whisper.cpp` will support more sampling strategies.
+- No GPU support (yet)
 
 ## Another example
 
@@ -234,7 +239,8 @@ in about half a minute on a MacBook M1 Pro, using `medium.en` model:
 ```java
 $ ./main -m models/ggml-medium.en.bin -f samples/gb1.wav -t 8
 
-whisper_model_load: loading model from 'models/ggml-medium.en.bin'
+whisper_init_from_file: loading model from 'models/ggml-medium.en.bin'
+whisper_model_load: loading model
 whisper_model_load: n_vocab       = 51864
 whisper_model_load: n_audio_ctx   = 1500
 whisper_model_load: n_audio_state = 1024
@@ -247,55 +253,60 @@ whisper_model_load: n_text_layer  = 24
 whisper_model_load: n_mels        = 80
 whisper_model_load: f16           = 1
 whisper_model_load: type          = 4
-whisper_model_load: mem_required  = 2610.00 MB
+whisper_model_load: mem required  = 1720.00 MB (+   43.00 MB per decoder)
+whisper_model_load: kv self size  =   42.00 MB
+whisper_model_load: kv cross size =  140.62 MB
 whisper_model_load: adding 1607 extra tokens
-whisper_model_load: ggml ctx size = 1644.97 MB
-whisper_model_load: memory size =   182.62 MB
-whisper_model_load: model size  =  1462.12 MB
+whisper_model_load: model ctx     = 1462.35 MB
+whisper_model_load: model size    = 1462.12 MB
 
-main: processing 'samples/gb1.wav' (3179750 samples, 198.7 sec), 8 threads, lang = en, task = transcribe, timestamps = 1 ...
+system_info: n_threads = 8 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 |
 
-[00:00.000 --> 00:08.000]   My fellow Americans, this day has brought terrible news and great sadness to our country.
-[00:08.000 --> 00:17.000]   At nine o'clock this morning, Mission Control in Houston lost contact with our Space Shuttle Columbia.
-[00:17.000 --> 00:23.000]   A short time later, debris was seen falling from the skies above Texas.
-[00:23.000 --> 00:29.000]   The Columbia's lost. There are no survivors.
-[00:29.000 --> 00:32.000]   On board was a crew of seven.
-[00:32.000 --> 00:39.000]   Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark,
-[00:39.000 --> 00:48.000]   Captain David Brown, Commander William McCool, Dr. Kultna Shavla, and Ilan Ramon,
-[00:48.000 --> 00:52.000]   a colonel in the Israeli Air Force.
-[00:52.000 --> 00:58.000]   These men and women assumed great risk in the service to all humanity.
-[00:58.000 --> 01:03.000]   In an age when space flight has come to seem almost routine,
-[01:03.000 --> 01:07.000]   it is easy to overlook the dangers of travel by rocket
-[01:07.000 --> 01:12.000]   and the difficulties of navigating the fierce outer atmosphere of the Earth.
-[01:12.000 --> 01:18.000]   These astronauts knew the dangers, and they faced them willingly,
-[01:18.000 --> 01:23.000]   knowing they had a high and noble purpose in life.
-[01:23.000 --> 01:31.000]   Because of their courage and daring and idealism, we will miss them all the more.
-[01:31.000 --> 01:36.000]   All Americans today are thinking as well of the families of these men and women
-[01:36.000 --> 01:40.000]   who have been given this sudden shock and grief.
-[01:40.000 --> 01:45.000]   You're not alone. Our entire nation grieves with you,
-[01:45.000 --> 01:52.000]   and those you love will always have the respect and gratitude of this country.
-[01:52.000 --> 01:56.000]   The cause in which they died will continue.
-[01:56.000 --> 02:04.000]   Mankind is led into the darkness beyond our world by the inspiration of discovery
-[02:04.000 --> 02:11.000]   and the longing to understand. Our journey into space will go on.
-[02:11.000 --> 02:16.000]   In the skies today, we saw destruction and tragedy.
-[02:16.000 --> 02:22.000]   Yet farther than we can see, there is comfort and hope.
-[02:22.000 --> 02:29.000]   In the words of the prophet Isaiah, "Lift your eyes and look to the heavens
-[02:29.000 --> 02:35.000]   who created all these. He who brings out the starry hosts one by one
-[02:35.000 --> 02:39.000]   and calls them each by name."
-[02:39.000 --> 02:46.000]   Because of His great power and mighty strength, not one of them is missing.
-[02:46.000 --> 02:55.000]   The same Creator who names the stars also knows the names of the seven souls we mourn today.
-[02:55.000 --> 03:01.000]   The crew of the shuttle Columbia did not return safely to earth,
-[03:01.000 --> 03:05.000]   yet we can pray that all are safely home.
-[03:05.000 --> 03:13.000]   May God bless the grieving families, and may God continue to bless America.
-[03:13.000 --> 03:41.000]   Audio
+main: processing 'samples/gb1.wav' (3179750 samples, 198.7 sec), 8 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
 
 
-whisper_print_timings:     load time =   575.92 ms
-whisper_print_timings:      mel time =   230.60 ms
-whisper_print_timings:   sample time =    73.19 ms
-whisper_print_timings:   encode time = 19552.61 ms / 814.69 ms per layer
-whisper_print_timings:   decode time = 13249.96 ms / 552.08 ms per layer
-whisper_print_timings:    total time = 33686.27 ms
+[00:00:00.000 --> 00:00:08.000]   My fellow Americans, this day has brought terrible news and great sadness to our country.
+[00:00:08.000 --> 00:00:17.000]   At nine o'clock this morning, Mission Control in Houston lost contact with our Space Shuttle Columbia.
+[00:00:17.000 --> 00:00:23.000]   A short time later, debris was seen falling from the skies above Texas.
+[00:00:23.000 --> 00:00:29.000]   The Columbia's lost. There are no survivors.
+[00:00:29.000 --> 00:00:32.000]   On board was a crew of seven.
+[00:00:32.000 --> 00:00:39.000]   Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark,
+[00:00:39.000 --> 00:00:48.000]   Captain David Brown, Commander William McCool, Dr. Kultna Shavla, and Ilan Ramon,
+[00:00:48.000 --> 00:00:52.000]   a colonel in the Israeli Air Force.
+[00:00:52.000 --> 00:00:58.000]   These men and women assumed great risk in the service to all humanity.
+[00:00:58.000 --> 00:01:03.000]   In an age when space flight has come to seem almost routine,
+[00:01:03.000 --> 00:01:07.000]   it is easy to overlook the dangers of travel by rocket
+[00:01:07.000 --> 00:01:12.000]   and the difficulties of navigating the fierce outer atmosphere of the Earth.
+[00:01:12.000 --> 00:01:18.000]   These astronauts knew the dangers, and they faced them willingly,
+[00:01:18.000 --> 00:01:23.000]   knowing they had a high and noble purpose in life.
+[00:01:23.000 --> 00:01:31.000]   Because of their courage and daring and idealism, we will miss them all the more.
+[00:01:31.000 --> 00:01:36.000]   All Americans today are thinking as well of the families of these men and women
+[00:01:36.000 --> 00:01:40.000]   who have been given this sudden shock and grief.
+[00:01:40.000 --> 00:01:45.000]   You're not alone. Our entire nation grieves with you,
+[00:01:45.000 --> 00:01:52.000]   and those you love will always have the respect and gratitude of this country.
+[00:01:52.000 --> 00:01:56.000]   The cause in which they died will continue.
+[00:01:56.000 --> 00:02:04.000]   Mankind is led into the darkness beyond our world by the inspiration of discovery
+[00:02:04.000 --> 00:02:11.000]   and the longing to understand. Our journey into space will go on.
+[00:02:11.000 --> 00:02:16.000]   In the skies today, we saw destruction and tragedy.
+[00:02:16.000 --> 00:02:22.000]   Yet farther than we can see, there is comfort and hope.
+[00:02:22.000 --> 00:02:29.000]   In the words of the prophet Isaiah, "Lift your eyes and look to the heavens
+[00:02:29.000 --> 00:02:35.000]   who created all these. He who brings out the starry hosts one by one
+[00:02:35.000 --> 00:02:39.000]   and calls them each by name."
+[00:02:39.000 --> 00:02:46.000]   Because of His great power and mighty strength, not one of them is missing.
+[00:02:46.000 --> 00:02:55.000]   The same Creator who names the stars also knows the names of the seven souls we mourn today.
+[00:02:55.000 --> 00:03:01.000]   The crew of the shuttle Columbia did not return safely to earth,
+[00:03:01.000 --> 00:03:05.000]   yet we can pray that all are safely home.
+[00:03:05.000 --> 00:03:13.000]   May God bless the grieving families, and may God continue to bless America.
+[00:03:13.000 --> 00:03:19.000]   [Silence]
+
+
+whisper_print_timings:     fallbacks =   1 p /   0 h
+whisper_print_timings:     load time =   569.03 ms
+whisper_print_timings:      mel time =   146.85 ms
+whisper_print_timings:   sample time =   238.66 ms /   553 runs (    0.43 ms per run)
+whisper_print_timings:   encode time = 18665.10 ms /     9 runs ( 2073.90 ms per run)
+whisper_print_timings:   decode time = 13090.93 ms /   549 runs (   23.85 ms per run)
+whisper_print_timings:    total time = 32733.52 ms
 ```
 </details>
 
@@ -306,6 +317,7 @@ The [stream](examples/stream) tool samples the audio every half a second and run
 More info is available in [issue #10](https://github.com/ggerganov/whisper.cpp/issues/10).
 
 ```java
+make stream
 ./stream -m ./models/ggml-base.en.bin -t 8 --step 500 --length 5000
 ```
 
@@ -320,14 +332,14 @@ to highlight words with high or low confidence:
 
 ## Controlling the length of the generated text segments (experimental)
 
-For example, to limit the line length to a maximum of 16 characters, simply add `-ml 16`: 
+For example, to limit the line length to a maximum of 16 characters, simply add `-ml 16`:
 
 ```java
 ./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 16
 
 whisper_model_load: loading model from './models/ggml-base.en.bin'
 ...
-system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | 
+system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |
 
 main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
 
@@ -351,7 +363,7 @@ The `--max-len` argument can be used to obtain word-level timestamps. Simply use
 
 whisper_model_load: loading model from './models/ggml-base.en.bin'
 ...
-system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | 
+system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |
 
 main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
 
@@ -447,12 +459,13 @@ or manually from here:
 For more details, see the conversion script [models/convert-pt-to-ggml.py](models/convert-pt-to-ggml.py) or the README
 in [models](models).
 
-## Bindings
+## [Bindings](https://github.com/ggerganov/whisper.cpp/discussions/categories/bindings)
 
-- [X] Rust: [tazz4843/whisper-rs](https://github.com/tazz4843/whisper-rs)
-- [X] Objective-C / Swift: [ggerganov/whisper.spm](https://github.com/ggerganov/whisper.spm)
-- [X] Javascript: [bindings/javascript](bindings/javascript)
-- [ ] Python: soon
+- [X] Rust: [tazz4843/whisper-rs](https://github.com/tazz4843/whisper-rs) | [#310](https://github.com/ggerganov/whisper.cpp/discussions/310)
+- [X] Javascript: [bindings/javascript](bindings/javascript) | [#309](https://github.com/ggerganov/whisper.cpp/discussions/309)
+- [X] Go: [bindings/go](bindings/go) | [#312](https://github.com/ggerganov/whisper.cpp/discussions/312)
+- [X] Objective-C / Swift: [ggerganov/whisper.spm](https://github.com/ggerganov/whisper.spm) | [#313](https://github.com/ggerganov/whisper.cpp/discussions/313)
+- [ ] Python: soon | [WIP](https://github.com/ggerganov/whisper.cpp/issues/9)
 
 ## Examples
 
@@ -467,6 +480,7 @@ Some of the examples are even ported to run in the browser using WebAssembly. Ch
 | [command](examples/command) | [command.wasm](examples/command.wasm) | Basic voice assistant example for receiving voice commands from the mic |
 | [talk](examples/talk) | [talk.wasm](examples/talk.wasm) | Talk with a GPT-2 bot |
 | [whisper.objc](examples/whisper.objc) | | iOS mobile application using whisper.cpp |
+| [whisper.swiftui](examples/whisper.swiftui) | | SwiftUI iOS / macOS application using whisper.cpp |
 | [whisper.android](examples/whisper.android) | | Android mobile application using whisper.cpp |
 | [whisper.nvim](examples/whisper.nvim) | | Speech-to-text plugin for Neovim |
 | [generate-karaoke.sh](examples/generate-karaoke.sh) | | Helper script to easily [generate a karaoke video](https://youtu.be/uj7hVta4blM) of raw audio capture |

bindings/go/.gitignore

@@ -0,0 +1,2 @@
+build
+models

bindings/go/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2022 David Thorpe
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

bindings/go/Makefile

@@ -0,0 +1,38 @@
+BUILD_DIR := build
+MODELS_DIR := models
+EXAMPLES_DIR := $(wildcard examples/*)
+INCLUDE_PATH := $(abspath ../..)
+LIBRARY_PATH := $(abspath ../..)
+
+all: clean whisper examples
+
+whisper: mkdir
+	@echo Build whisper
+	@${MAKE} -C ../.. libwhisper.a
+
+test: model-small whisper modtidy
+	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} go test -v .
+	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} go test -v ./pkg/whisper/...
+
+examples: $(EXAMPLES_DIR)
+
+model-small: mkdir examples/go-model-download
+	@${BUILD_DIR}/go-model-download -out models ggml-small.en.bin
+
+$(EXAMPLES_DIR): mkdir whisper modtidy
+	@echo Build example $(notdir $@)
+	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} go build ${BUILD_FLAGS} -o ${BUILD_DIR}/$(notdir $@) ./$@
+
+mkdir:
+	@echo Mkdir ${BUILD_DIR}
+	@install -d ${BUILD_DIR}
+	@echo Mkdir ${MODELS_DIR}
+	@install -d ${MODELS_DIR}
+
+modtidy:
+	@go mod tidy
+
+clean: 
+	@echo Clean
+	@rm -fr $(BUILD_DIR)
+	@go clean

bindings/go/README.md

@@ -0,0 +1,100 @@
+# Go bindings for Whisper
+
+This package provides Go bindings for whisper.cpp. They have been tested on:
+
+  * Darwin (OS X) 12.6 on x64_64
+  * Debian Linux on arm64
+  * Fedora Linux on x86_64
+
+The "low level" bindings are in the `bindings/go` directory and there is a more
+Go-style package in the `bindings/go/pkg/whisper` directory. The most simple usage
+is as follows:
+
+```go
+import (
+	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+)
+
+func main() {
+	var modelpath string // Path to the model
+	var samples []float32 // Samples to process
+
+	// Load the model
+	model, err := whisper.New(modelpath)
+	if err != nil {
+		panic(err)
+	}
+	defer model.Close()
+
+	// Process samples
+	context, err := model.NewContext()
+	if err != nil {
+		panic(err)
+	}
+	if err := context.Process(samples, nil); err != nil {
+		return err
+	}
+
+	// Print out the results
+	for {
+		segment, err := context.NextSegment()
+		if err != nil {
+			break
+		}
+		fmt.Printf("[%6s->%6s] %s\n", segment.Start, segment.End, segment.Text)
+	}
+}
+```
+
+## Building & Testing
+
+In order to build, you need to have the Go compiler installed. You can get it from [here](https://golang.org/dl/). Run the tests with:
+
+```bash
+git clone https://github.com/ggerganov/whisper.cpp.git
+cd whisper.cpp/bindings/go
+make test
+```
+
+This will compile a static `libwhisper.a` in a `build` folder, download a model file, then run the tests. To build the examples:
+
+```bash
+make examples
+```
+
+The examples are placed in the `build` directory. Once built, you can download all the models with the following command:
+
+```bash
+./build/go-model-download -out models
+```
+
+And you can then test a model against samples with the following command:
+
+```bash
+./build/go-whisper -model models/ggml-tiny.en.bin samples/jfk.wav 
+```
+
+## Using the bindings
+
+To use the bindings in your own software,
+
+  1. Import `github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper` (or `github.com/ggerganov/whisper.cpp/bindings/go` into your package;
+  2. Compile `libwhisper.a` (you can use `make whisper` in the `bindings/go` directory);
+  3. Link your go binary against whisper by setting the environment variables `C_INCLUDE_PATH` and `LIBRARY_PATH`
+     to point to the `whisper.h` file directory and `libwhisper.a` file directory respectively.
+
+Look at the `Makefile` in the `bindings/go` directory for an example.
+
+The API Documentation:
+
+  * https://pkg.go.dev/github.com/ggerganov/whisper.cpp/bindings/go
+  * https://pkg.go.dev/github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper
+
+Getting help:
+
+  * Follow the discussion for the go bindings [here](https://github.com/ggerganov/whisper.cpp/discussions/312)
+
+## License
+
+The license for the Go bindings is the same as the license for the rest of the whisper.cpp project, which is the MIT License. See the `LICENSE` file for more details.
+

bindings/go/doc.go

@@ -0,0 +1,5 @@
+/*
+github.com/ggerganov/whisper.cpp/bindings/go
+provides a speech-to-text service bindings for the Go programming language.
+*/
+package whisper

bindings/go/examples/go-model-download/context.go

@@ -0,0 +1,30 @@
+package main
+
+import (
+	"context"
+	"os"
+	"os/signal"
+)
+
+// ContextForSignal returns a context object which is cancelled when a signal
+// is received. It returns nil if no signal parameter is provided
+func ContextForSignal(signals ...os.Signal) context.Context {
+	if len(signals) == 0 {
+		return nil
+	}
+
+	ch := make(chan os.Signal)
+	ctx, cancel := context.WithCancel(context.Background())
+
+	// Send message on channel when signal received
+	signal.Notify(ch, signals...)
+
+	// When any signal received, call cancel
+	go func() {
+		<-ch
+		cancel()
+	}()
+
+	// Return success
+	return ctx
+}

bindings/go/examples/go-model-download/main.go

@@ -0,0 +1,208 @@
+package main
+
+import (
+	"context"
+	"flag"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"os"
+	"path/filepath"
+	"syscall"
+	"time"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// CONSTANTS
+
+const (
+	srcUrl  = "https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main" // The location of the models
+	srcExt  = ".bin"                                                               // Filename extension
+	bufSize = 1024 * 64                                                            // Size of the buffer used for downloading the model
+)
+
+var (
+	// The models which will be downloaded, if no model is specified as an argument
+	modelNames = []string{"ggml-tiny.en", "ggml-tiny", "ggml-base.en", "ggml-base", "ggml-small.en", "ggml-small", "ggml-medium.en", "ggml-medium", "ggml-large-v1", "ggml-large"}
+)
+
+var (
+	// The output folder. When not set, use current working directory.
+	flagOut = flag.String("out", "", "Output folder")
+
+	// HTTP timeout parameter - will timeout if takes longer than this to download a model
+	flagTimeout = flag.Duration("timeout", 30*time.Minute, "HTTP timeout")
+
+	// Quiet parameter - will not print progress if set
+	flagQuiet = flag.Bool("quiet", false, "Quiet mode")
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// MAIN
+
+func main() {
+	flag.Usage = func() {
+		name := filepath.Base(flag.CommandLine.Name())
+		fmt.Fprintf(flag.CommandLine.Output(), "Usage: %s [options] <model>\n\n", name)
+		flag.PrintDefaults()
+	}
+	flag.Parse()
+
+	// Get output path
+	out, err := GetOut()
+	if err != nil {
+		fmt.Fprintln(os.Stderr, "Error:", err)
+		os.Exit(-1)
+	}
+
+	// Create context which quits on SIGINT or SIGQUIT
+	ctx := ContextForSignal(os.Interrupt, syscall.SIGQUIT)
+
+	// Progress filehandle
+	progress := os.Stdout
+	if *flagQuiet {
+		progress, err = os.Open(os.DevNull)
+		if err != nil {
+			fmt.Fprintln(os.Stderr, "Error:", err)
+			os.Exit(-1)
+		}
+		defer progress.Close()
+	}
+
+	// Download models - exit on error or interrupt
+	for _, model := range GetModels() {
+		url, err := URLForModel(model)
+		if err != nil {
+			fmt.Fprintln(os.Stderr, "Error:", err)
+			continue
+		} else if path, err := Download(ctx, progress, url, out); err == nil || err == io.EOF {
+			continue
+		} else if err == context.Canceled {
+			os.Remove(path)
+			fmt.Fprintln(progress, "\nInterrupted")
+			break
+		} else if err == context.DeadlineExceeded {
+			os.Remove(path)
+			fmt.Fprintln(progress, "Timeout downloading model")
+			continue
+		} else {
+			os.Remove(path)
+			fmt.Fprintln(os.Stderr, "Error:", err)
+			break
+		}
+	}
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+// GetOut returns the path to the output directory
+func GetOut() (string, error) {
+	if *flagOut == "" {
+		return os.Getwd()
+	}
+	if info, err := os.Stat(*flagOut); err != nil {
+		return "", err
+	} else if !info.IsDir() {
+		return "", fmt.Errorf("not a directory: %s", info.Name())
+	} else {
+		return *flagOut, nil
+	}
+}
+
+// GetModels returns the list of models to download
+func GetModels() []string {
+	if flag.NArg() == 0 {
+		return modelNames
+	} else {
+		return flag.Args()
+	}
+}
+
+// URLForModel returns the URL for the given model on huggingface.co
+func URLForModel(model string) (string, error) {
+	if filepath.Ext(model) != srcExt {
+		model += srcExt
+	}
+	url, err := url.Parse(srcUrl)
+	if err != nil {
+		return "", err
+	} else {
+		url.Path = filepath.Join(url.Path, model)
+	}
+	return url.String(), nil
+}
+
+// Download downloads the model from the given URL to the given output directory
+func Download(ctx context.Context, p io.Writer, model, out string) (string, error) {
+	// Create HTTP client
+	client := http.Client{
+		Timeout: *flagTimeout,
+	}
+
+	// Initiate the download
+	req, err := http.NewRequest("GET", model, nil)
+	if err != nil {
+		return "", err
+	}
+	resp, err := client.Do(req)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		return "", fmt.Errorf("%s: %s", model, resp.Status)
+	}
+
+	// If output file exists and is the same size as the model, skip
+	path := filepath.Join(out, filepath.Base(model))
+	if info, err := os.Stat(path); err == nil && info.Size() == resp.ContentLength {
+		fmt.Fprintln(p, "Skipping", model, "as it already exists")
+		return "", nil
+	}
+
+	// Create file
+	w, err := os.Create(path)
+	if err != nil {
+		return "", err
+	}
+	defer w.Close()
+
+	// Report
+	fmt.Fprintln(p, "Downloading", model, "to", out)
+
+	// Progressively download the model
+	data := make([]byte, bufSize)
+	count, pct := int64(0), int64(0)
+	ticker := time.NewTicker(5 * time.Second)
+	for {
+		select {
+		case <-ctx.Done():
+			// Cancelled, return error
+			return path, ctx.Err()
+		case <-ticker.C:
+			pct = DownloadReport(p, pct, count, resp.ContentLength)
+		default:
+			// Read body
+			n, err := resp.Body.Read(data)
+			if err != nil {
+				DownloadReport(p, pct, count, resp.ContentLength)
+				return path, err
+			} else if m, err := w.Write(data[:n]); err != nil {
+				return path, err
+			} else {
+				count += int64(m)
+			}
+		}
+	}
+}
+
+// Report periodically reports the download progress when percentage changes
+func DownloadReport(w io.Writer, pct, count, total int64) int64 {
+	pct_ := count * 100 / total
+	if pct_ > pct {
+		fmt.Fprintf(w, "  ...%d MB written (%d%%)\n", count/1e6, pct_)
+	}
+	return pct_
+}

bindings/go/examples/go-whisper/color.go

@@ -0,0 +1,22 @@
+package main
+
+import "fmt"
+
+///////////////////////////////////////////////////////////////////////////////
+// CONSTANTS
+
+const (
+	Reset     = "\033[0m"
+	RGBPrefix = "\033[38;5;" // followed by RGB values in decimal format separated by colons
+	RGBSuffix = "m"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+// Colorize text with RGB values, from 0 to 23
+func Colorize(text string, v int) string {
+	// https://en.wikipedia.org/wiki/ANSI_escape_code#8-bit
+	// Grayscale colors are in the range 232-255
+	return RGBPrefix + fmt.Sprint(v%24+232) + RGBSuffix + text + Reset
+}

bindings/go/examples/go-whisper/flags.go

@@ -0,0 +1,156 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"strings"
+	"time"
+
+	// Packages
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+type Flags struct {
+	*flag.FlagSet
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// LIFECYCLE
+
+func NewFlags(name string, args []string) (*Flags, error) {
+	flags := &Flags{
+		FlagSet: flag.NewFlagSet(name, flag.ContinueOnError),
+	}
+
+	// Register the command line arguments
+	registerFlags(flags)
+
+	// Parse command line
+	if err := flags.Parse(args); err != nil {
+		return nil, err
+	}
+
+	// Return success
+	return flags, nil
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+func (flags *Flags) GetModel() string {
+	return flags.Lookup("model").Value.String()
+}
+
+func (flags *Flags) GetLanguage() string {
+	return flags.Lookup("language").Value.String()
+}
+
+func (flags *Flags) IsTranslate() bool {
+	return flags.Lookup("translate").Value.(flag.Getter).Get().(bool)
+}
+
+func (flags *Flags) GetOffset() time.Duration {
+	return flags.Lookup("offset").Value.(flag.Getter).Get().(time.Duration)
+}
+
+func (flags *Flags) GetDuration() time.Duration {
+	return flags.Lookup("duration").Value.(flag.Getter).Get().(time.Duration)
+}
+
+func (flags *Flags) GetThreads() uint {
+	return flags.Lookup("threads").Value.(flag.Getter).Get().(uint)
+}
+
+func (flags *Flags) GetOut() string {
+	return strings.ToLower(flags.Lookup("out").Value.String())
+}
+
+func (flags *Flags) IsSpeedup() bool {
+	return flags.Lookup("speedup").Value.String() == "true"
+}
+
+func (flags *Flags) IsTokens() bool {
+	return flags.Lookup("tokens").Value.String() == "true"
+}
+
+func (flags *Flags) IsColorize() bool {
+	return flags.Lookup("colorize").Value.String() == "true"
+}
+
+func (flags *Flags) GetMaxLen() uint {
+	return flags.Lookup("max-len").Value.(flag.Getter).Get().(uint)
+}
+
+func (flags *Flags) GetMaxTokens() uint {
+	return flags.Lookup("max-tokens").Value.(flag.Getter).Get().(uint)
+}
+
+func (flags *Flags) GetWordThreshold() float32 {
+	return float32(flags.Lookup("word-thold").Value.(flag.Getter).Get().(float64))
+}
+
+func (flags *Flags) SetParams(context whisper.Context) error {
+	if lang := flags.GetLanguage(); lang != "" && lang != "auto" {
+		fmt.Fprintf(flags.Output(), "Setting language to %q\n", lang)
+		if err := context.SetLanguage(lang); err != nil {
+			return err
+		}
+	}
+	if flags.IsTranslate() && context.IsMultilingual() {
+		fmt.Fprintf(flags.Output(), "Setting translate to true\n")
+		context.SetTranslate(true)
+	}
+	if offset := flags.GetOffset(); offset != 0 {
+		fmt.Fprintf(flags.Output(), "Setting offset to %v\n", offset)
+		context.SetOffset(offset)
+	}
+	if duration := flags.GetDuration(); duration != 0 {
+		fmt.Fprintf(flags.Output(), "Setting duration to %v\n", duration)
+		context.SetDuration(duration)
+	}
+	if flags.IsSpeedup() {
+		fmt.Fprintf(flags.Output(), "Setting speedup to true\n")
+		context.SetSpeedup(true)
+	}
+	if threads := flags.GetThreads(); threads != 0 {
+		fmt.Fprintf(flags.Output(), "Setting threads to %d\n", threads)
+		context.SetThreads(threads)
+	}
+	if max_len := flags.GetMaxLen(); max_len != 0 {
+		fmt.Fprintf(flags.Output(), "Setting max_segment_length to %d\n", max_len)
+		context.SetMaxSegmentLength(max_len)
+	}
+	if max_tokens := flags.GetMaxTokens(); max_tokens != 0 {
+		fmt.Fprintf(flags.Output(), "Setting max_tokens to %d\n", max_tokens)
+		context.SetMaxTokensPerSegment(max_tokens)
+	}
+	if word_threshold := flags.GetWordThreshold(); word_threshold != 0 {
+		fmt.Fprintf(flags.Output(), "Setting word_threshold to %f\n", word_threshold)
+		context.SetTokenThreshold(word_threshold)
+	}
+
+	// Return success
+	return nil
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PRIVATE METHODS
+
+func registerFlags(flag *Flags) {
+	flag.String("model", "", "Path to the model file")
+	flag.String("language", "", "Spoken language")
+	flag.Bool("translate", false, "Translate from source language to english")
+	flag.Duration("offset", 0, "Time offset")
+	flag.Duration("duration", 0, "Duration of audio to process")
+	flag.Uint("threads", 0, "Number of threads to use")
+	flag.Bool("speedup", false, "Enable speedup")
+	flag.Uint("max-len", 0, "Maximum segment length in characters")
+	flag.Uint("max-tokens", 0, "Maximum tokens per segment")
+	flag.Float64("word-thold", 0, "Maximum segment score")
+	flag.Bool("tokens", false, "Display tokens")
+	flag.Bool("colorize", false, "Colorize tokens")
+	flag.String("out", "", "Output format (srt, none or leave as empty string)")
+}

bindings/go/examples/go-whisper/main.go

@@ -0,0 +1,43 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"path/filepath"
+
+	// Packages
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+)
+
+func main() {
+	flags, err := NewFlags(filepath.Base(os.Args[0]), os.Args[1:])
+	if err == flag.ErrHelp {
+		os.Exit(0)
+	} else if err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
+	} else if flags.GetModel() == "" {
+		fmt.Fprintln(os.Stderr, "Use -model flag to specify which model file to use")
+		os.Exit(1)
+	} else if flags.NArg() == 0 {
+		fmt.Fprintln(os.Stderr, "No input files specified")
+		os.Exit(1)
+	}
+
+	// Load model
+	model, err := whisper.New(flags.GetModel())
+	if err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
+	}
+	defer model.Close()
+
+	// Process files
+	for _, filename := range flags.Args() {
+		if err := Process(model, filename, flags); err != nil {
+			fmt.Fprintln(os.Stderr, err)
+			continue
+		}
+	}
+}

bindings/go/examples/go-whisper/process.go

@@ -0,0 +1,132 @@
+package main
+
+import (
+	"fmt"
+	"io"
+	"os"
+	"time"
+
+	// Package imports
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+	wav "github.com/go-audio/wav"
+)
+
+func Process(model whisper.Model, path string, flags *Flags) error {
+	var data []float32
+
+	// Create processing context
+	context, err := model.NewContext()
+	if err != nil {
+		return err
+	}
+
+	// Set the parameters
+	if err := flags.SetParams(context); err != nil {
+		return err
+	}
+
+	fmt.Printf("\n%s\n", context.SystemInfo())
+
+	// Open the file
+	fmt.Fprintf(flags.Output(), "Loading %q\n", path)
+	fh, err := os.Open(path)
+	if err != nil {
+		return err
+	}
+	defer fh.Close()
+
+	// Decode the WAV file - load the full buffer
+	dec := wav.NewDecoder(fh)
+	if buf, err := dec.FullPCMBuffer(); err != nil {
+		return err
+	} else if dec.SampleRate != whisper.SampleRate {
+		return fmt.Errorf("unsupported sample rate: %d", dec.SampleRate)
+	} else if dec.NumChans != 1 {
+		return fmt.Errorf("unsupported number of channels: %d", dec.NumChans)
+	} else {
+		data = buf.AsFloat32Buffer().Data
+	}
+
+	// Segment callback when -tokens is specified
+	var cb whisper.SegmentCallback
+	if flags.IsTokens() {
+		cb = func(segment whisper.Segment) {
+			fmt.Fprintf(flags.Output(), "%02d [%6s->%6s] ", segment.Num, segment.Start.Truncate(time.Millisecond), segment.End.Truncate(time.Millisecond))
+			for _, token := range segment.Tokens {
+				if flags.IsColorize() && context.IsText(token) {
+					fmt.Fprint(flags.Output(), Colorize(token.Text, int(token.P*24.0)), " ")
+				} else {
+					fmt.Fprint(flags.Output(), token.Text, " ")
+				}
+			}
+			fmt.Fprintln(flags.Output(), "")
+			fmt.Fprintln(flags.Output(), "")
+		}
+	}
+
+	// Process the data
+	fmt.Fprintf(flags.Output(), "  ...processing %q\n", path)
+	context.ResetTimings()
+	if err := context.Process(data, cb); err != nil {
+		return err
+	}
+
+	context.PrintTimings()
+
+	// Print out the results
+	switch {
+	case flags.GetOut() == "srt":
+		return OutputSRT(os.Stdout, context)
+	case flags.GetOut() == "none":
+		return nil
+	default:
+		return Output(os.Stdout, context, flags.IsColorize())
+	}
+}
+
+// Output text as SRT file
+func OutputSRT(w io.Writer, context whisper.Context) error {
+	n := 1
+	for {
+		segment, err := context.NextSegment()
+		if err == io.EOF {
+			return nil
+		} else if err != nil {
+			return err
+		}
+		fmt.Fprintln(w, n)
+		fmt.Fprintln(w, srtTimestamp(segment.Start), " --> ", srtTimestamp(segment.End))
+		fmt.Fprintln(w, segment.Text)
+		fmt.Fprintln(w, "")
+		n++
+	}
+}
+
+// Output text to terminal
+func Output(w io.Writer, context whisper.Context, colorize bool) error {
+	for {
+		segment, err := context.NextSegment()
+		if err == io.EOF {
+			return nil
+		} else if err != nil {
+			return err
+		}
+		fmt.Fprintf(w, "[%6s->%6s]", segment.Start.Truncate(time.Millisecond), segment.End.Truncate(time.Millisecond))
+		if colorize {
+			for _, token := range segment.Tokens {
+				if !context.IsText(token) {
+					continue
+				}
+				fmt.Fprint(w, " ", Colorize(token.Text, int(token.P*24.0)))
+			}
+			fmt.Fprint(w, "\n")
+		} else {
+			fmt.Fprintln(w, " ", segment.Text)
+		}
+	}
+}
+
+// Return srtTimestamp
+func srtTimestamp(t time.Duration) string {
+	return fmt.Sprintf("%02d:%02d:%02d,%03d", t/time.Hour, (t%time.Hour)/time.Minute, (t%time.Minute)/time.Second, (t%time.Second)/time.Millisecond)
+}

bindings/go/go.mod

@@ -0,0 +1,16 @@
+module github.com/ggerganov/whisper.cpp/bindings/go
+
+go 1.19
+
+require (
+	github.com/go-audio/wav v1.1.0
+	github.com/stretchr/testify v1.8.1
+)
+
+require (
+	github.com/davecgh/go-spew v1.1.1 // indirect
+	github.com/go-audio/audio v1.0.0 // indirect
+	github.com/go-audio/riff v1.0.0 // indirect
+	github.com/pmezard/go-difflib v1.0.0 // indirect
+	gopkg.in/yaml.v3 v3.0.1 // indirect
+)

bindings/go/go.sum

@@ -0,0 +1,23 @@
+github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
+github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/go-audio/audio v1.0.0 h1:zS9vebldgbQqktK4H0lUqWrG8P0NxCJVqcj7ZpNnwd4=
+github.com/go-audio/audio v1.0.0/go.mod h1:6uAu0+H2lHkwdGsAY+j2wHPNPpPoeg5AaEFh9FlA+Zs=
+github.com/go-audio/riff v1.0.0 h1:d8iCGbDvox9BfLagY94fBynxSPHO80LmZCaOsmKxokA=
+github.com/go-audio/riff v1.0.0/go.mod h1:l3cQwc85y79NQFCRB7TiPoNiaijp6q8Z0Uv38rVG498=
+github.com/go-audio/wav v1.1.0 h1:jQgLtbqBzY7G+BM8fXF7AHUk1uHUviWS4X39d5rsL2g=
+github.com/go-audio/wav v1.1.0/go.mod h1:mpe9qfwbScEbkd8uybLuIpTgHyrISw/OTuvjUW2iGtE=
+github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
+github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
+github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
+github.com/stretchr/objx v0.4.0/go.mod h1:YvHI0jy2hoMjB+UWwv71VJQ9isScKT/TqJzVSSt89Yw=
+github.com/stretchr/objx v0.5.0/go.mod h1:Yh+to48EsGEfYuaHDzXPcE3xhTkx73EhmCGUpEOglKo=
+github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
+github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU=
+github.com/stretchr/testify v1.8.1 h1:w7B6lhMri9wdJUVmEZPGGhZzrYTPvgJArz7wNPgYKsk=
+github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
+gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

bindings/go/params.go

@@ -0,0 +1,165 @@
+package whisper
+
+import (
+	"fmt"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// CGO
+
+/*
+#include <whisper.h>
+*/
+import "C"
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+func (p *Params) SetTranslate(v bool) {
+	p.translate = toBool(v)
+}
+
+func (p *Params) SetNoContext(v bool) {
+	p.no_context = toBool(v)
+}
+
+func (p *Params) SetSingleSegment(v bool) {
+	p.single_segment = toBool(v)
+}
+
+func (p *Params) SetPrintSpecial(v bool) {
+	p.print_special = toBool(v)
+}
+
+func (p *Params) SetPrintProgress(v bool) {
+	p.print_progress = toBool(v)
+}
+
+func (p *Params) SetPrintRealtime(v bool) {
+	p.print_realtime = toBool(v)
+}
+
+func (p *Params) SetPrintTimestamps(v bool) {
+	p.print_timestamps = toBool(v)
+}
+
+func (p *Params) SetSpeedup(v bool) {
+	p.speed_up = toBool(v)
+}
+
+// Set language id
+func (p *Params) SetLanguage(lang int) error {
+	if lang == -1 {
+		p.language = nil
+		return nil
+	}
+	str := C.whisper_lang_str(C.int(lang))
+	if str == nil {
+		return ErrInvalidLanguage
+	} else {
+		p.language = str
+	}
+	return nil
+}
+
+// Get language id
+func (p *Params) Language() int {
+	if p.language == nil {
+		return -1
+	}
+	return int(C.whisper_lang_id(p.language))
+}
+
+// Threads available
+func (p *Params) Threads() int {
+	return int(p.n_threads)
+}
+
+// Set number of threads to use
+func (p *Params) SetThreads(threads int) {
+	p.n_threads = C.int(threads)
+}
+
+// Set start offset in ms
+func (p *Params) SetOffset(offset_ms int) {
+	p.offset_ms = C.int(offset_ms)
+}
+
+// Set audio duration to process in ms
+func (p *Params) SetDuration(duration_ms int) {
+	p.duration_ms = C.int(duration_ms)
+}
+
+// Set timestamp token probability threshold (~0.01)
+func (p *Params) SetTokenThreshold(t float32) {
+	p.thold_pt = C.float(t)
+}
+
+// Set timestamp token sum probability threshold (~0.01)
+func (p *Params) SetTokenSumThreshold(t float32) {
+	p.thold_ptsum = C.float(t)
+}
+
+// Set max segment length in characters
+func (p *Params) SetMaxSegmentLength(n int) {
+	p.max_len = C.int(n)
+}
+
+// Set max tokens per segment (0 = no limit)
+func (p *Params) SetMaxTokensPerSegment(n int) {
+	p.max_tokens = C.int(n)
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PRIVATE METHODS
+
+func toBool(v bool) C.bool {
+	if v {
+		return C.bool(true)
+	}
+	return C.bool(false)
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// STRINGIFY
+
+func (p *Params) String() string {
+	str := "<whisper.params"
+	str += fmt.Sprintf(" strategy=%v", p.strategy)
+	str += fmt.Sprintf(" n_threads=%d", p.n_threads)
+	if p.language != nil {
+		str += fmt.Sprintf(" language=%s", C.GoString(p.language))
+	}
+	str += fmt.Sprintf(" n_max_text_ctx=%d", p.n_max_text_ctx)
+	str += fmt.Sprintf(" offset_ms=%d", p.offset_ms)
+	str += fmt.Sprintf(" duration_ms=%d", p.duration_ms)
+	if p.translate {
+		str += " translate"
+	}
+	if p.no_context {
+		str += " no_context"
+	}
+	if p.single_segment {
+		str += " single_segment"
+	}
+	if p.print_special {
+		str += " print_special"
+	}
+	if p.print_progress {
+		str += " print_progress"
+	}
+	if p.print_realtime {
+		str += " print_realtime"
+	}
+	if p.print_timestamps {
+		str += " print_timestamps"
+	}
+	if p.token_timestamps {
+		str += " token_timestamps"
+	}
+	if p.speed_up {
+		str += " speed_up"
+	}
+
+	return str + ">"
+}

bindings/go/pkg/whisper/consts.go

@@ -0,0 +1,28 @@
+package whisper
+
+import (
+	"errors"
+
+	// Bindings
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// ERRORS
+
+var (
+	ErrUnableToLoadModel    = errors.New("unable to load model")
+	ErrInternalAppError     = errors.New("internal application error")
+	ErrProcessingFailed     = errors.New("processing failed")
+	ErrUnsupportedLanguage  = errors.New("unsupported language")
+	ErrModelNotMultilingual = errors.New("model is not multilingual")
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// CONSTANTS
+
+// SampleRate is the sample rate of the audio data.
+const SampleRate = whisper.SampleRate
+
+// SampleBits is the number of bytes per sample.
+const SampleBits = whisper.SampleBits

bindings/go/pkg/whisper/context.go

@@ -0,0 +1,290 @@
+package whisper
+
+import (
+	"fmt"
+	"io"
+	"runtime"
+	"strings"
+	"time"
+
+	// Bindings
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+type context struct {
+	n      int
+	model  *model
+	params whisper.Params
+}
+
+// Make sure context adheres to the interface
+var _ Context = (*context)(nil)
+
+///////////////////////////////////////////////////////////////////////////////
+// LIFECYCLE
+
+func newContext(model *model, params whisper.Params) (Context, error) {
+	context := new(context)
+	context.model = model
+	context.params = params
+
+	// Return success
+	return context, nil
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+// Set the language to use for speech recognition.
+func (context *context) SetLanguage(lang string) error {
+	if context.model.ctx == nil {
+		return ErrInternalAppError
+	}
+	if !context.model.IsMultilingual() {
+		return ErrModelNotMultilingual
+	}
+
+	if lang == "auto" {
+		context.params.SetLanguage(-1)
+	} else if id := context.model.ctx.Whisper_lang_id(lang); id < 0 {
+		return ErrUnsupportedLanguage
+	} else if err := context.params.SetLanguage(id); err != nil {
+		return err
+	}
+	// Return success
+	return nil
+}
+
+func (context *context) IsMultilingual() bool {
+	return context.model.IsMultilingual()
+}
+
+// Get language
+func (context *context) Language() string {
+	id := context.params.Language()
+	if id == -1 {
+		return "auto"
+	}
+	return whisper.Whisper_lang_str(context.params.Language())
+}
+
+// Set translate flag
+func (context *context) SetTranslate(v bool) {
+	context.params.SetTranslate(v)
+}
+
+// Set speedup flag
+func (context *context) SetSpeedup(v bool) {
+	context.params.SetSpeedup(v)
+}
+
+// Set number of threads to use
+func (context *context) SetThreads(v uint) {
+	context.params.SetThreads(int(v))
+}
+
+// Set time offset
+func (context *context) SetOffset(v time.Duration) {
+	context.params.SetOffset(int(v.Milliseconds()))
+}
+
+// Set duration of audio to process
+func (context *context) SetDuration(v time.Duration) {
+	context.params.SetOffset(int(v.Milliseconds()))
+}
+
+// Set timestamp token probability threshold (~0.01)
+func (context *context) SetTokenThreshold(t float32) {
+	context.params.SetTokenThreshold(t)
+}
+
+// Set timestamp token sum probability threshold (~0.01)
+func (context *context) SetTokenSumThreshold(t float32) {
+	context.params.SetTokenSumThreshold(t)
+}
+
+// Set max segment length in characters
+func (context *context) SetMaxSegmentLength(n uint) {
+	context.params.SetMaxSegmentLength(int(n))
+}
+
+// Set max tokens per segment (0 = no limit)
+func (context *context) SetMaxTokensPerSegment(n uint) {
+	context.params.SetMaxTokensPerSegment(int(n))
+}
+
+// ResetTimings resets the mode timings. Should be called before processing
+func (context *context) ResetTimings() {
+	context.model.ctx.Whisper_reset_timings()
+}
+
+// PrintTimings prints the model timings to stdout.
+func (context *context) PrintTimings() {
+	context.model.ctx.Whisper_print_timings()
+}
+
+// SystemInfo returns the system information
+func (context *context) SystemInfo() string {
+	return fmt.Sprintf("system_info: n_threads = %d / %d | %s\n",
+		context.params.Threads(),
+		runtime.NumCPU(),
+		whisper.Whisper_print_system_info(),
+	)
+}
+
+// Use mel data at offset_ms to try and auto-detect the spoken language
+// Make sure to call whisper_pcm_to_mel() or whisper_set_mel() first.
+// Returns the probabilities of all languages.
+func (context *context) WhisperLangAutoDetect(offset_ms int, n_threads int) ([]float32, error) {
+	langProbs, err := context.model.ctx.Whisper_lang_auto_detect(offset_ms, n_threads)
+	if err != nil {
+		return nil, err
+	}
+	return langProbs, nil
+}
+
+// Process new sample data and return any errors
+func (context *context) Process(data []float32, cb SegmentCallback) error {
+	if context.model.ctx == nil {
+		return ErrInternalAppError
+	}
+	// If the callback is defined then we force on single_segment mode
+	if cb != nil {
+		context.params.SetSingleSegment(true)
+	}
+
+	// We don't do parallel processing at the moment
+	processors := 0
+	if processors > 1 {
+		if err := context.model.ctx.Whisper_full_parallel(context.params, data, processors, nil, func(new int) {
+			if cb != nil {
+				num_segments := context.model.ctx.Whisper_full_n_segments()
+				s0 := num_segments - new
+				for i := s0; i < num_segments; i++ {
+					cb(toSegment(context.model.ctx, i))
+				}
+			}
+		}); err != nil {
+			return err
+		}
+	} else if err := context.model.ctx.Whisper_full(context.params, data, nil, func(new int) {
+		if cb != nil {
+			num_segments := context.model.ctx.Whisper_full_n_segments()
+			s0 := num_segments - new
+			for i := s0; i < num_segments; i++ {
+				cb(toSegment(context.model.ctx, i))
+			}
+		}
+	}); err != nil {
+		return err
+	}
+
+	// Return success
+	return nil
+}
+
+// Return the next segment of tokens
+func (context *context) NextSegment() (Segment, error) {
+	if context.model.ctx == nil {
+		return Segment{}, ErrInternalAppError
+	}
+	if context.n >= context.model.ctx.Whisper_full_n_segments() {
+		return Segment{}, io.EOF
+	}
+
+	// Populate result
+	result := toSegment(context.model.ctx, context.n)
+
+	// Increment the cursor
+	context.n++
+
+	// Return success
+	return result, nil
+}
+
+// Test for text tokens
+func (context *context) IsText(t Token) bool {
+	switch {
+	case context.IsBEG(t):
+		return false
+	case context.IsSOT(t):
+		return false
+	case whisper.Token(t.Id) >= context.model.ctx.Whisper_token_eot():
+		return false
+	case context.IsPREV(t):
+		return false
+	case context.IsSOLM(t):
+		return false
+	case context.IsNOT(t):
+		return false
+	default:
+		return true
+	}
+}
+
+// Test for "begin" token
+func (context *context) IsBEG(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_beg()
+}
+
+// Test for "start of transcription" token
+func (context *context) IsSOT(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_sot()
+}
+
+// Test for "end of transcription" token
+func (context *context) IsEOT(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_eot()
+}
+
+// Test for "start of prev" token
+func (context *context) IsPREV(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_prev()
+}
+
+// Test for "start of lm" token
+func (context *context) IsSOLM(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_solm()
+}
+
+// Test for "No timestamps" token
+func (context *context) IsNOT(t Token) bool {
+	return whisper.Token(t.Id) == context.model.ctx.Whisper_token_not()
+}
+
+// Test for token associated with a specific language
+func (context *context) IsLANG(t Token, lang string) bool {
+	if id := context.model.ctx.Whisper_lang_id(lang); id >= 0 {
+		return whisper.Token(t.Id) == context.model.ctx.Whisper_token_lang(id)
+	} else {
+		return false
+	}
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PRIVATE METHODS
+
+func toSegment(ctx *whisper.Context, n int) Segment {
+	return Segment{
+		Num:    n,
+		Text:   strings.TrimSpace(ctx.Whisper_full_get_segment_text(n)),
+		Start:  time.Duration(ctx.Whisper_full_get_segment_t0(n)) * time.Millisecond * 10,
+		End:    time.Duration(ctx.Whisper_full_get_segment_t1(n)) * time.Millisecond * 10,
+		Tokens: toTokens(ctx, n),
+	}
+}
+
+func toTokens(ctx *whisper.Context, n int) []Token {
+	result := make([]Token, ctx.Whisper_full_n_tokens(n))
+	for i := 0; i < len(result); i++ {
+		result[i] = Token{
+			Id:   int(ctx.Whisper_full_get_token_id(n, i)),
+			Text: strings.TrimSpace(ctx.Whisper_full_get_token_text(n, i)),
+			P:    ctx.Whisper_full_get_token_p(n, i),
+		}
+	}
+	return result
+}

bindings/go/pkg/whisper/context_test.go

@@ -0,0 +1,55 @@
+package whisper_test
+
+import (
+	"os"
+	"testing"
+
+	// Packages
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+	assert "github.com/stretchr/testify/assert"
+)
+
+const (
+	ModelPath  = "../../models/ggml-tiny.bin"
+	SamplePath = "../../samples/jfk.wav"
+)
+
+func Test_Whisper_000(t *testing.T) {
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	if _, err := os.Stat(SamplePath); os.IsNotExist(err) {
+		t.Skip("Skipping test, sample not found:", SamplePath)
+	}
+
+	// Load model
+	model, err := whisper.New(ModelPath)
+	assert.NoError(err)
+	assert.NotNil(model)
+	assert.NoError(model.Close())
+
+	t.Log("languages=", model.Languages())
+}
+
+func Test_Whisper_001(t *testing.T) {
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	if _, err := os.Stat(SamplePath); os.IsNotExist(err) {
+		t.Skip("Skipping test, sample not found:", SamplePath)
+	}
+
+	// Load model
+	model, err := whisper.New(ModelPath)
+	assert.NoError(err)
+	assert.NotNil(model)
+	defer model.Close()
+
+	// Get context for decoding
+	ctx, err := model.NewContext()
+	assert.NoError(err)
+	assert.NotNil(ctx)
+
+}

bindings/go/pkg/whisper/doc.go

@@ -0,0 +1,4 @@
+/*
+This is the higher-level speech-to-text whisper.cpp API for go
+*/
+package whisper

bindings/go/pkg/whisper/interface.go

@@ -0,0 +1,91 @@
+package whisper
+
+import (
+	"io"
+	"time"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+// SegmentCallback is the callback function for processing segments in real
+// time. It is called during the Process function
+type SegmentCallback func(Segment)
+
+// Model is the interface to a whisper model. Create a new model with the
+// function whisper.New(string)
+type Model interface {
+	io.Closer
+
+	// Return a new speech-to-text context.
+	NewContext() (Context, error)
+
+	// Return true if the model is multilingual.
+	IsMultilingual() bool
+
+	// Return all languages supported.
+	Languages() []string
+}
+
+// Context is the speach recognition context.
+type Context interface {
+	SetLanguage(string) error // Set the language to use for speech recognition, use "auto" for auto detect language.
+	SetTranslate(bool)        // Set translate flag
+	IsMultilingual() bool     // Return true if the model is multilingual.
+	Language() string         // Get language
+
+	SetOffset(time.Duration)      // Set offset
+	SetDuration(time.Duration)    // Set duration
+	SetThreads(uint)              // Set number of threads to use
+	SetSpeedup(bool)              // Set speedup flag
+	SetTokenThreshold(float32)    // Set timestamp token probability threshold
+	SetTokenSumThreshold(float32) // Set timestamp token sum probability threshold
+	SetMaxSegmentLength(uint)     // Set max segment length in characters
+	SetMaxTokensPerSegment(uint)  // Set max tokens per segment (0 = no limit)
+
+	// Process mono audio data and return any errors.
+	// If defined, newly generated segments are passed to the
+	// callback function during processing.
+	Process([]float32, SegmentCallback) error
+
+	// After process is called, return segments until the end of the stream
+	// is reached, when io.EOF is returned.
+	NextSegment() (Segment, error)
+
+	IsBEG(Token) bool          // Test for "begin" token
+	IsSOT(Token) bool          // Test for "start of transcription" token
+	IsEOT(Token) bool          // Test for "end of transcription" token
+	IsPREV(Token) bool         // Test for "start of prev" token
+	IsSOLM(Token) bool         // Test for "start of lm" token
+	IsNOT(Token) bool          // Test for "No timestamps" token
+	IsLANG(Token, string) bool // Test for token associated with a specific language
+	IsText(Token) bool         // Test for text token
+
+	// Timings
+	PrintTimings()
+	ResetTimings()
+
+	SystemInfo() string
+}
+
+// Segment is the text result of a speech recognition.
+type Segment struct {
+	// Segment Number
+	Num int
+
+	// Time beginning and end timestamps for the segment.
+	Start, End time.Duration
+
+	// The text of the segment.
+	Text string
+
+	// The tokens of the segment.
+	Tokens []Token
+}
+
+// Token is a text or special token
+type Token struct {
+	Id   int
+	Text string
+	P    float32
+}

bindings/go/pkg/whisper/model.go

@@ -0,0 +1,100 @@
+package whisper
+
+import (
+	"fmt"
+	"os"
+	"runtime"
+
+	// Bindings
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+type model struct {
+	path string
+	ctx  *whisper.Context
+}
+
+// Make sure model adheres to the interface
+var _ Model = (*model)(nil)
+
+///////////////////////////////////////////////////////////////////////////////
+// LIFECYCLE
+
+func New(path string) (Model, error) {
+	model := new(model)
+	if _, err := os.Stat(path); err != nil {
+		return nil, err
+	} else if ctx := whisper.Whisper_init(path); ctx == nil {
+		return nil, ErrUnableToLoadModel
+	} else {
+		model.ctx = ctx
+		model.path = path
+	}
+
+	// Return success
+	return model, nil
+}
+
+func (model *model) Close() error {
+	if model.ctx != nil {
+		model.ctx.Whisper_free()
+	}
+
+	// Release resources
+	model.ctx = nil
+
+	// Return success
+	return nil
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// STRINGIFY
+
+func (model *model) String() string {
+	str := "<whisper.model"
+	if model.ctx != nil {
+		str += fmt.Sprintf(" model=%q", model.path)
+	}
+	return str + ">"
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+// Return true if model is multilingual (language and translation options are supported)
+func (model *model) IsMultilingual() bool {
+	return model.ctx.Whisper_is_multilingual() != 0
+}
+
+// Return all recognized languages. Initially it is set to auto-detect
+func (model *model) Languages() []string {
+	result := make([]string, 0, whisper.Whisper_lang_max_id())
+	for i := 0; i < whisper.Whisper_lang_max_id(); i++ {
+		str := whisper.Whisper_lang_str(i)
+		if model.ctx.Whisper_lang_id(str) >= 0 {
+			result = append(result, str)
+		}
+	}
+	return result
+}
+
+func (model *model) NewContext() (Context, error) {
+	if model.ctx == nil {
+		return nil, ErrInternalAppError
+	}
+
+	// Create new context
+	params := model.ctx.Whisper_full_default_params(whisper.SAMPLING_GREEDY)
+	params.SetTranslate(false)
+	params.SetPrintSpecial(false)
+	params.SetPrintProgress(false)
+	params.SetPrintRealtime(false)
+	params.SetPrintTimestamps(false)
+	params.SetThreads(runtime.NumCPU())
+
+	// Return new context
+	return newContext(model, params)
+}

bindings/go/whisper.go

@@ -0,0 +1,409 @@
+package whisper
+
+import (
+	"errors"
+	"unsafe"
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// CGO
+
+/*
+#cgo LDFLAGS: -lwhisper -lm -lstdc++
+#cgo darwin LDFLAGS: -framework Accelerate
+#include <whisper.h>
+#include <stdlib.h>
+
+extern void callNewSegment(void* user_data, int new);
+extern bool callEncoderBegin(void* user_data);
+
+// Text segment callback
+// Called on every newly generated text segment
+// Use the whisper_full_...() functions to obtain the text segments
+static void whisper_new_segment_cb(struct whisper_context* ctx, int n_new, void* user_data) {
+    if(user_data != NULL && ctx != NULL) {
+        callNewSegment(user_data, n_new);
+    }
+}
+
+// Encoder begin callback
+// If not NULL, called before the encoder starts
+// If it returns false, the computation is aborted
+static bool whisper_encoder_begin_cb(struct whisper_context* ctx, void* user_data) {
+    if(user_data != NULL && ctx != NULL) {
+        return callEncoderBegin(user_data);
+    }
+    return false;
+}
+
+// Get default parameters and set callbacks
+static struct whisper_full_params whisper_full_default_params_cb(struct whisper_context* ctx, enum whisper_sampling_strategy strategy) {
+	struct whisper_full_params params = whisper_full_default_params(strategy);
+	params.new_segment_callback = whisper_new_segment_cb;
+	params.new_segment_callback_user_data = (void*)(ctx);
+	params.encoder_begin_callback = whisper_encoder_begin_cb;
+	params.encoder_begin_callback_user_data = (void*)(ctx);
+	return params;
+}
+*/
+import "C"
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+type (
+	Context          C.struct_whisper_context
+	Token            C.whisper_token
+	TokenData        C.struct_whisper_token_data
+	SamplingStrategy C.enum_whisper_sampling_strategy
+	Params           C.struct_whisper_full_params
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// GLOBALS
+
+const (
+	SAMPLING_GREEDY      SamplingStrategy = C.WHISPER_SAMPLING_GREEDY
+	SAMPLING_BEAM_SEARCH SamplingStrategy = C.WHISPER_SAMPLING_BEAM_SEARCH
+)
+
+const (
+	SampleRate = C.WHISPER_SAMPLE_RATE                 // Expected sample rate, samples per second
+	SampleBits = uint16(unsafe.Sizeof(C.float(0))) * 8 // Sample size in bits
+	NumFFT     = C.WHISPER_N_FFT
+	NumMEL     = C.WHISPER_N_MEL
+	HopLength  = C.WHISPER_HOP_LENGTH
+	ChunkSize  = C.WHISPER_CHUNK_SIZE
+)
+
+var (
+	ErrTokenizerFailed  = errors.New("whisper_tokenize failed")
+	ErrAutoDetectFailed = errors.New("whisper_lang_auto_detect failed")
+	ErrConversionFailed = errors.New("whisper_convert failed")
+	ErrInvalidLanguage  = errors.New("invalid language")
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// PUBLIC METHODS
+
+// Allocates all memory needed for the model and loads the model from the given file.
+// Returns NULL on failure.
+func Whisper_init(path string) *Context {
+	cPath := C.CString(path)
+	defer C.free(unsafe.Pointer(cPath))
+	if ctx := C.whisper_init_from_file(cPath); ctx != nil {
+		return (*Context)(ctx)
+	} else {
+		return nil
+	}
+}
+
+// Frees all memory allocated by the model.
+func (ctx *Context) Whisper_free() {
+	C.whisper_free((*C.struct_whisper_context)(ctx))
+}
+
+// Convert RAW PCM audio to log mel spectrogram.
+// The resulting spectrogram is stored inside the provided whisper context.
+func (ctx *Context) Whisper_pcm_to_mel(data []float32, threads int) error {
+	if C.whisper_pcm_to_mel((*C.struct_whisper_context)(ctx), (*C.float)(&data[0]), C.int(len(data)), C.int(threads)) == 0 {
+		return nil
+	} else {
+		return ErrConversionFailed
+	}
+}
+
+// This can be used to set a custom log mel spectrogram inside the provided whisper context.
+// Use this instead of whisper_pcm_to_mel() if you want to provide your own log mel spectrogram.
+// n_mel must be 80
+func (ctx *Context) Whisper_set_mel(data []float32, n_mel int) error {
+	if C.whisper_set_mel((*C.struct_whisper_context)(ctx), (*C.float)(&data[0]), C.int(len(data)), C.int(n_mel)) == 0 {
+		return nil
+	} else {
+		return ErrConversionFailed
+	}
+}
+
+// Run the Whisper encoder on the log mel spectrogram stored inside the provided whisper context.
+// Make sure to call whisper_pcm_to_mel() or whisper_set_mel() first.
+// offset can be used to specify the offset of the first frame in the spectrogram.
+func (ctx *Context) Whisper_encode(offset, threads int) error {
+	if C.whisper_encode((*C.struct_whisper_context)(ctx), C.int(offset), C.int(threads)) == 0 {
+		return nil
+	} else {
+		return ErrConversionFailed
+	}
+}
+
+// Run the Whisper decoder to obtain the logits and probabilities for the next token.
+// Make sure to call whisper_encode() first.
+// tokens + n_tokens is the provided context for the decoder.
+// n_past is the number of tokens to use from previous decoder calls.
+func (ctx *Context) Whisper_decode(tokens []Token, past, threads int) error {
+	if C.whisper_decode((*C.struct_whisper_context)(ctx), (*C.whisper_token)(&tokens[0]), C.int(len(tokens)), C.int(past), C.int(threads)) == 0 {
+		return nil
+	} else {
+		return ErrConversionFailed
+	}
+}
+
+// Convert the provided text into tokens. The tokens pointer must be large enough to hold the resulting tokens.
+// Returns the number of tokens on success
+func (ctx *Context) Whisper_tokenize(text string, tokens []Token) (int, error) {
+	cText := C.CString(text)
+	defer C.free(unsafe.Pointer(cText))
+	if n := C.whisper_tokenize((*C.struct_whisper_context)(ctx), cText, (*C.whisper_token)(&tokens[0]), C.int(len(tokens))); n >= 0 {
+		return int(n), nil
+	} else {
+		return 0, ErrTokenizerFailed
+	}
+}
+
+// Return the id of the specified language, returns -1 if not found
+// Examples:
+//
+//	"de" -> 2
+//	"german" -> 2
+func (ctx *Context) Whisper_lang_id(lang string) int {
+	return int(C.whisper_lang_id(C.CString(lang)))
+}
+
+// Largest language id (i.e. number of available languages - 1)
+func Whisper_lang_max_id() int {
+	return int(C.whisper_lang_max_id())
+}
+
+// Return the short string of the specified language id (e.g. 2 -> "de"),
+// returns empty string if not found
+func Whisper_lang_str(id int) string {
+	return C.GoString(C.whisper_lang_str(C.int(id)))
+}
+
+// Use mel data at offset_ms to try and auto-detect the spoken language
+// Make sure to call whisper_pcm_to_mel() or whisper_set_mel() first.
+// Returns the probabilities of all languages.
+// ref: https://github.com/openai/whisper/blob/main/whisper/decoding.py#L18-L69
+func (ctx *Context) Whisper_lang_auto_detect(offset_ms, n_threads int) ([]float32, error) {
+	probs := make([]float32, Whisper_lang_max_id()+1)
+	if n := int(C.whisper_lang_auto_detect((*C.struct_whisper_context)(ctx), C.int(offset_ms), C.int(n_threads), (*C.float)(&probs[0]))); n < 0 {
+		return nil, ErrAutoDetectFailed
+	} else {
+		return probs, nil
+	}
+}
+
+func (ctx *Context) Whisper_n_len() int {
+	return int(C.whisper_n_len((*C.struct_whisper_context)(ctx)))
+}
+
+func (ctx *Context) Whisper_n_vocab() int {
+	return int(C.whisper_n_vocab((*C.struct_whisper_context)(ctx)))
+}
+
+func (ctx *Context) Whisper_n_text_ctx() int {
+	return int(C.whisper_n_text_ctx((*C.struct_whisper_context)(ctx)))
+}
+
+func (ctx *Context) Whisper_n_audio_ctx() int {
+	return int(C.whisper_n_audio_ctx((*C.struct_whisper_context)(ctx)))
+}
+
+func (ctx *Context) Whisper_is_multilingual() int {
+	return int(C.whisper_is_multilingual((*C.struct_whisper_context)(ctx)))
+}
+
+// The probabilities for the next token
+//func (ctx *Whisper_context) Whisper_get_probs() []float32 {
+//	return (*[1 << 30]float32)(unsafe.Pointer(C.whisper_get_probs((*C.struct_whisper_context)(ctx))))[:ctx.Whisper_n_vocab()]
+//}
+
+// Token Id -> String. Uses the vocabulary in the provided context
+func (ctx *Context) Whisper_token_to_str(token Token) string {
+	return C.GoString(C.whisper_token_to_str((*C.struct_whisper_context)(ctx), C.whisper_token(token)))
+}
+
+// Special tokens
+func (ctx *Context) Whisper_token_eot() Token {
+	return Token(C.whisper_token_eot((*C.struct_whisper_context)(ctx)))
+}
+
+// Special tokens
+func (ctx *Context) Whisper_token_sot() Token {
+	return Token(C.whisper_token_sot((*C.struct_whisper_context)(ctx)))
+}
+
+// Special tokens
+func (ctx *Context) Whisper_token_prev() Token {
+	return Token(C.whisper_token_prev((*C.struct_whisper_context)(ctx)))
+}
+
+// Special tokens
+func (ctx *Context) Whisper_token_solm() Token {
+	return Token(C.whisper_token_solm((*C.struct_whisper_context)(ctx)))
+}
+
+// Special tokens
+func (ctx *Context) Whisper_token_not() Token {
+	return Token(C.whisper_token_not((*C.struct_whisper_context)(ctx)))
+}
+
+// Special tokens
+func (ctx *Context) Whisper_token_beg() Token {
+	return Token(C.whisper_token_beg((*C.struct_whisper_context)(ctx)))
+}
+
+// Special tokens
+func (ctx *Context) Whisper_token_lang(lang_id int) Token {
+	return Token(C.whisper_token_lang((*C.struct_whisper_context)(ctx), C.int(lang_id)))
+}
+
+// Task tokens
+func Whisper_token_translate() Token {
+	return Token(C.whisper_token_translate())
+}
+
+// Task tokens
+func Whisper_token_transcribe() Token {
+	return Token(C.whisper_token_transcribe())
+}
+
+// Performance information
+func (ctx *Context) Whisper_print_timings() {
+	C.whisper_print_timings((*C.struct_whisper_context)(ctx))
+}
+
+// Performance information
+func (ctx *Context) Whisper_reset_timings() {
+	C.whisper_reset_timings((*C.struct_whisper_context)(ctx))
+}
+
+// Print system information
+func Whisper_print_system_info() string {
+	return C.GoString(C.whisper_print_system_info())
+}
+
+// Return default parameters for a strategy
+func (ctx *Context) Whisper_full_default_params(strategy SamplingStrategy) Params {
+	// Get default parameters
+	return Params(C.whisper_full_default_params_cb((*C.struct_whisper_context)(ctx), C.enum_whisper_sampling_strategy(strategy)))
+}
+
+// Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
+// Uses the specified decoding strategy to obtain the text.
+func (ctx *Context) Whisper_full(params Params, samples []float32, encoderBeginCallback func() bool, newSegmentCallback func(int)) error {
+	registerEncoderBeginCallback(ctx, encoderBeginCallback)
+	registerNewSegmentCallback(ctx, newSegmentCallback)
+	defer registerEncoderBeginCallback(ctx, nil)
+	defer registerNewSegmentCallback(ctx, nil)
+	if C.whisper_full((*C.struct_whisper_context)(ctx), (C.struct_whisper_full_params)(params), (*C.float)(&samples[0]), C.int(len(samples))) == 0 {
+		return nil
+	} else {
+		return ErrConversionFailed
+	}
+}
+
+// Split the input audio in chunks and process each chunk separately using whisper_full()
+// It seems this approach can offer some speedup in some cases.
+// However, the transcription accuracy can be worse at the beginning and end of each chunk.
+func (ctx *Context) Whisper_full_parallel(params Params, samples []float32, processors int, encoderBeginCallback func() bool, newSegmentCallback func(int)) error {
+	registerEncoderBeginCallback(ctx, encoderBeginCallback)
+	registerNewSegmentCallback(ctx, newSegmentCallback)
+	defer registerEncoderBeginCallback(ctx, nil)
+	defer registerNewSegmentCallback(ctx, nil)
+
+	if C.whisper_full_parallel((*C.struct_whisper_context)(ctx), (C.struct_whisper_full_params)(params), (*C.float)(&samples[0]), C.int(len(samples)), C.int(processors)) == 0 {
+		return nil
+	} else {
+		return ErrConversionFailed
+	}
+}
+
+// Number of generated text segments.
+// A segment can be a few words, a sentence, or even a paragraph.
+func (ctx *Context) Whisper_full_n_segments() int {
+	return int(C.whisper_full_n_segments((*C.struct_whisper_context)(ctx)))
+}
+
+// Get the start and end time of the specified segment.
+func (ctx *Context) Whisper_full_get_segment_t0(segment int) int64 {
+	return int64(C.whisper_full_get_segment_t0((*C.struct_whisper_context)(ctx), C.int(segment)))
+}
+
+// Get the start and end time of the specified segment.
+func (ctx *Context) Whisper_full_get_segment_t1(segment int) int64 {
+	return int64(C.whisper_full_get_segment_t1((*C.struct_whisper_context)(ctx), C.int(segment)))
+}
+
+// Get the text of the specified segment.
+func (ctx *Context) Whisper_full_get_segment_text(segment int) string {
+	return C.GoString(C.whisper_full_get_segment_text((*C.struct_whisper_context)(ctx), C.int(segment)))
+}
+
+// Get number of tokens in the specified segment.
+func (ctx *Context) Whisper_full_n_tokens(segment int) int {
+	return int(C.whisper_full_n_tokens((*C.struct_whisper_context)(ctx), C.int(segment)))
+}
+
+// Get the token text of the specified token index in the specified segment.
+func (ctx *Context) Whisper_full_get_token_text(segment int, token int) string {
+	return C.GoString(C.whisper_full_get_token_text((*C.struct_whisper_context)(ctx), C.int(segment), C.int(token)))
+}
+
+// Get the token of the specified token index in the specified segment.
+func (ctx *Context) Whisper_full_get_token_id(segment int, token int) Token {
+	return Token(C.whisper_full_get_token_id((*C.struct_whisper_context)(ctx), C.int(segment), C.int(token)))
+}
+
+// Get token data for the specified token in the specified segment.
+// This contains probabilities, timestamps, etc.
+func (ctx *Context) whisper_full_get_token_data(segment int, token int) TokenData {
+	return TokenData(C.whisper_full_get_token_data((*C.struct_whisper_context)(ctx), C.int(segment), C.int(token)))
+}
+
+// Get the probability of the specified token in the specified segment.
+func (ctx *Context) Whisper_full_get_token_p(segment int, token int) float32 {
+	return float32(C.whisper_full_get_token_p((*C.struct_whisper_context)(ctx), C.int(segment), C.int(token)))
+}
+
+///////////////////////////////////////////////////////////////////////////////
+// CALLBACKS
+
+var (
+	cbNewSegment   = make(map[unsafe.Pointer]func(int))
+	cbEncoderBegin = make(map[unsafe.Pointer]func() bool)
+)
+
+func registerNewSegmentCallback(ctx *Context, fn func(int)) {
+	if fn == nil {
+		delete(cbNewSegment, unsafe.Pointer(ctx))
+	} else {
+		cbNewSegment[unsafe.Pointer(ctx)] = fn
+	}
+}
+
+func registerEncoderBeginCallback(ctx *Context, fn func() bool) {
+	if fn == nil {
+		delete(cbEncoderBegin, unsafe.Pointer(ctx))
+	} else {
+		cbEncoderBegin[unsafe.Pointer(ctx)] = fn
+	}
+}
+
+//export callNewSegment
+func callNewSegment(user_data unsafe.Pointer, new C.int) {
+	if fn, ok := cbNewSegment[user_data]; ok {
+		fn(int(new))
+	}
+}
+
+//export callEncoderBegin
+func callEncoderBegin(user_data unsafe.Pointer) C.bool {
+	if fn, ok := cbEncoderBegin[user_data]; ok {
+		if fn() {
+			return C.bool(true)
+		} else {
+			return C.bool(false)
+		}
+	}
+	return true
+}

bindings/go/whisper_test.go

@@ -0,0 +1,113 @@
+package whisper_test
+
+import (
+	"os"
+	"runtime"
+	"testing"
+	"time"
+
+	// Packages
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go"
+	wav "github.com/go-audio/wav"
+	assert "github.com/stretchr/testify/assert"
+)
+
+const (
+	ModelPath  = "models/ggml-small.en.bin"
+	SamplePath = "samples/jfk.wav"
+)
+
+func Test_Whisper_000(t *testing.T) {
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	ctx := whisper.Whisper_init(ModelPath)
+	assert.NotNil(ctx)
+	ctx.Whisper_free()
+}
+
+func Test_Whisper_001(t *testing.T) {
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	if _, err := os.Stat(SamplePath); os.IsNotExist(err) {
+		t.Skip("Skipping test, sample not found:", SamplePath)
+	}
+
+	// Open samples
+	fh, err := os.Open(SamplePath)
+	assert.NoError(err)
+	defer fh.Close()
+
+	// Read samples
+	d := wav.NewDecoder(fh)
+	buf, err := d.FullPCMBuffer()
+	assert.NoError(err)
+
+	// Run whisper
+	ctx := whisper.Whisper_init(ModelPath)
+	assert.NotNil(ctx)
+	defer ctx.Whisper_free()
+	params := ctx.Whisper_full_default_params(whisper.SAMPLING_GREEDY)
+	data := buf.AsFloat32Buffer().Data
+	err = ctx.Whisper_full(params, data, nil, nil)
+	assert.NoError(err)
+
+	// Print out tokens
+	num_segments := ctx.Whisper_full_n_segments()
+	assert.GreaterOrEqual(num_segments, 1)
+	for i := 0; i < num_segments; i++ {
+		str := ctx.Whisper_full_get_segment_text(i)
+		assert.NotEmpty(str)
+		t0 := time.Duration(ctx.Whisper_full_get_segment_t0(i)) * time.Millisecond
+		t1 := time.Duration(ctx.Whisper_full_get_segment_t1(i)) * time.Millisecond
+		t.Logf("[%6s->%-6s] %q", t0, t1, str)
+	}
+}
+
+func Test_Whisper_002(t *testing.T) {
+	assert := assert.New(t)
+	for i := 0; i < whisper.Whisper_lang_max_id(); i++ {
+		str := whisper.Whisper_lang_str(i)
+		assert.NotEmpty(str)
+		t.Log(str)
+	}
+}
+
+func Test_Whisper_003(t *testing.T) {
+	threads := runtime.NumCPU()
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	if _, err := os.Stat(SamplePath); os.IsNotExist(err) {
+		t.Skip("Skipping test, sample not found:", SamplePath)
+	}
+
+	// Open samples
+	fh, err := os.Open(SamplePath)
+	assert.NoError(err)
+	defer fh.Close()
+
+	// Read samples
+	d := wav.NewDecoder(fh)
+	buf, err := d.FullPCMBuffer()
+	assert.NoError(err)
+
+	// Make the model
+	ctx := whisper.Whisper_init(ModelPath)
+	assert.NotNil(ctx)
+	defer ctx.Whisper_free()
+
+	// Get MEL
+	assert.NoError(ctx.Whisper_pcm_to_mel(buf.AsFloat32Buffer().Data, threads))
+
+	// Get Languages
+	languages, err := ctx.Whisper_lang_auto_detect(0, threads)
+	assert.NoError(err)
+	for i, p := range languages {
+		t.Logf("%s: %f", whisper.Whisper_lang_str(i), p)
+	}
+}

bindings/javascript/emscripten.cpp

@@ -20,7 +20,7 @@ struct whisper_context * g_context;
 EMSCRIPTEN_BINDINGS(whisper) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         if (g_context == nullptr) {
-            g_context = whisper_init(path_model.c_str());
+            g_context = whisper_init_from_file(path_model.c_str());
             if (g_context != nullptr) {
                 return true;
             } else {

bindings/javascript/package.json

@@ -1,6 +1,6 @@
 {
   "name": "whisper.cpp",
-  "version": "1.0.4",
+  "version": "1.2.0",
   "description": "Whisper speech recognition",
   "main": "whisper.js",
   "scripts": {

cmake/DefaultTargetOptions.cmake

@@ -0,0 +1,17 @@
+# Set the default compile features and properties for a target.
+
+if (NOT TARGET)
+    message(FATAL_ERROR "TARGET not set before including DefaultTargetOptions")
+endif()
+
+target_compile_features(${TARGET}
+    PRIVATE
+        cxx_std_11
+    )
+
+set_target_properties(${TARGET}
+    PROPERTIES
+        EXPORT_COMPILE_COMMANDS ON
+        RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/bin"
+        INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/lib"
+)

examples/CMakeLists.txt

@@ -24,6 +24,8 @@ if (EMSCRIPTEN)
     add_subdirectory(command.wasm)
     add_subdirectory(talk.wasm)
     add_subdirectory(bench.wasm)
+elseif(CMAKE_JS_VERSION)
+    add_subdirectory(addon.node)
 else()
     add_subdirectory(main)
     add_subdirectory(stream)

examples/addon.node/.gitignore

@@ -0,0 +1,3 @@
+.idea
+node_modules
+build

examples/addon.node/CMakeLists.txt

@@ -0,0 +1,26 @@
+set(TARGET whisper-addon)
+
+# Base settings
+#==================================================================
+# env var supported by cmake-js
+add_definitions(-DNAPI_VERSION=4)
+include_directories(${CMAKE_JS_INC})
+#==================================================================
+
+add_library(${TARGET} SHARED ${CMAKE_JS_SRC} addon.cpp)
+set_target_properties(${TARGET} PROPERTIES PREFIX "" SUFFIX ".node")
+
+include(DefaultTargetOptions)
+
+# Include N-API wrappers
+#==================================================================
+execute_process(COMMAND node -p "require('node-addon-api').include"
+        WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
+        OUTPUT_VARIABLE NODE_ADDON_API_DIR
+        )
+string(REPLACE "\n" "" NODE_ADDON_API_DIR ${NODE_ADDON_API_DIR})
+string(REPLACE "\"" "" NODE_ADDON_API_DIR ${NODE_ADDON_API_DIR})
+target_include_directories(${TARGET} PRIVATE ${NODE_ADDON_API_DIR})
+#==================================================================
+
+target_link_libraries(${TARGET} ${CMAKE_JS_LIB} whisper ${CMAKE_THREAD_LIBS_INIT})

examples/addon.node/README.md

@@ -0,0 +1,37 @@
+# addon
+
+This is an addon demo that can **perform whisper model reasoning in `node` and `electron` environments**, based on [cmake-js](https://github.com/cmake-js/cmake-js).
+It can be used as a reference for using the whisper.cpp project in other node projects.
+
+## Install
+
+```shell
+npm install
+```
+
+## Compile
+
+Make sure it is in the project root directory and compiled with make-js.
+
+```shell
+npx cmake-js compile -T whisper-addon
+```
+
+For Electron addon and cmake-js options, you can see [cmake-js](https://github.com/cmake-js/cmake-js) and make very few configuration changes.
+
+> Such as appointing special cmake path:
+> ```shell
+> npx cmake-js compile -c 'xxx/cmake' -T whisper-addon
+> ```
+
+## Run
+
+```shell
+cd examples/addon.node
+
+node index.js --language='language' --model='model-path' --fname_inp='file-path'
+```
+
+Because this is a simple Demo, only the above parameters are set in the node environment.
+
+Other parameters can also be specified in the node environment.

examples/addon.node/addon.cpp

@@ -0,0 +1,421 @@
+#include <string>
+#include <thread>
+#include <vector>
+#include <cmath>
+
+#include "napi.h"
+
+#define DR_WAV_IMPLEMENTATION
+#include "dr_wav.h"
+
+#include "whisper.h"
+
+struct whisper_params {
+    int32_t n_threads    = std::min(4, (int32_t) std::thread::hardware_concurrency());
+    int32_t n_processors = 1;
+    int32_t offset_t_ms  = 0;
+    int32_t offset_n     = 0;
+    int32_t duration_ms  = 0;
+    int32_t max_context  = -1;
+    int32_t max_len      = 0;
+    int32_t best_of      = 5;
+    int32_t beam_size    = -1;
+
+    float word_thold    = 0.01f;
+    float entropy_thold = 2.4f;
+    float logprob_thold = -1.0f;
+
+    bool speed_up       = false;
+    bool translate      = false;
+    bool diarize        = false;
+    bool output_txt     = false;
+    bool output_vtt     = false;
+    bool output_srt     = false;
+    bool output_wts     = false;
+    bool output_csv     = false;
+    bool print_special  = false;
+    bool print_colors   = false;
+    bool print_progress = false;
+    bool no_timestamps  = false;
+
+    std::string language = "en";
+    std::string prompt;
+    std::string model    = "../../ggml-large.bin";
+
+    std::vector<std::string> fname_inp = {};
+    std::vector<std::string> fname_outp = {};
+};
+
+struct whisper_print_user_data {
+    const whisper_params * params;
+
+    const std::vector<std::vector<float>> * pcmf32s;
+};
+
+//  500 -> 00:05.000
+// 6000 -> 01:00.000
+std::string to_timestamp(int64_t t, bool comma = false) {
+    int64_t msec = t * 10;
+    int64_t hr = msec / (1000 * 60 * 60);
+    msec = msec - hr * (1000 * 60 * 60);
+    int64_t min = msec / (1000 * 60);
+    msec = msec - min * (1000 * 60);
+    int64_t sec = msec / 1000;
+    msec = msec - sec * 1000;
+
+    char buf[32];
+    snprintf(buf, sizeof(buf), "%02d:%02d:%02d%s%03d", (int) hr, (int) min, (int) sec, comma ? "," : ".", (int) msec);
+
+    return std::string(buf);
+}
+
+int timestamp_to_sample(int64_t t, int n_samples) {
+    return std::max(0, std::min((int) n_samples - 1, (int) ((t*WHISPER_SAMPLE_RATE)/100)));
+}
+
+void whisper_print_segment_callback(struct whisper_context * ctx, int n_new, void * user_data) {
+    const auto & params  = *((whisper_print_user_data *) user_data)->params;
+    const auto & pcmf32s = *((whisper_print_user_data *) user_data)->pcmf32s;
+
+    const int n_segments = whisper_full_n_segments(ctx);
+
+    std::string speaker = "";
+
+    int64_t t0;
+    int64_t t1;
+
+    // print the last n_new segments
+    const int s0 = n_segments - n_new;
+
+    if (s0 == 0) {
+        printf("\n");
+    }
+
+    for (int i = s0; i < n_segments; i++) {
+        if (!params.no_timestamps || params.diarize) {
+            t0 = whisper_full_get_segment_t0(ctx, i);
+            t1 = whisper_full_get_segment_t1(ctx, i);
+        }
+
+        if (!params.no_timestamps) {
+            printf("[%s --> %s]  ", to_timestamp(t0).c_str(), to_timestamp(t1).c_str());
+        }
+
+        if (params.diarize && pcmf32s.size() == 2) {
+            const int64_t n_samples = pcmf32s[0].size();
+
+            const int64_t is0 = timestamp_to_sample(t0, n_samples);
+            const int64_t is1 = timestamp_to_sample(t1, n_samples);
+
+            double energy0 = 0.0f;
+            double energy1 = 0.0f;
+
+            for (int64_t j = is0; j < is1; j++) {
+                energy0 += fabs(pcmf32s[0][j]);
+                energy1 += fabs(pcmf32s[1][j]);
+            }
+
+            if (energy0 > 1.1*energy1) {
+                speaker = "(speaker 0)";
+            } else if (energy1 > 1.1*energy0) {
+                speaker = "(speaker 1)";
+            } else {
+                speaker = "(speaker ?)";
+            }
+
+            //printf("is0 = %lld, is1 = %lld, energy0 = %f, energy1 = %f, %s\n", is0, is1, energy0, energy1, speaker.c_str());
+        }
+
+        // colorful print bug
+        //
+        const char * text = whisper_full_get_segment_text(ctx, i);
+        printf("%s%s", speaker.c_str(), text);
+
+
+        // with timestamps or speakers: each segment on new line
+        if (!params.no_timestamps || params.diarize) {
+            printf("\n");
+        }
+
+        fflush(stdout);
+    }
+}
+
+int run(whisper_params &params, std::vector<std::vector<std::string>> &result) {
+
+    if (params.fname_inp.empty()) {
+        fprintf(stderr, "error: no input files specified\n");
+        return 2;
+    }
+
+    if (params.language != "auto" && whisper_lang_id(params.language.c_str()) == -1) {
+        fprintf(stderr, "error: unknown language '%s'\n", params.language.c_str());
+        exit(0);
+    }
+
+    // whisper init
+
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
+
+    if (ctx == nullptr) {
+        fprintf(stderr, "error: failed to initialize whisper context\n");
+        return 3;
+    }
+
+    // initial prompt
+    std::vector<whisper_token> prompt_tokens;
+
+    if (!params.prompt.empty()) {
+        prompt_tokens.resize(1024);
+        prompt_tokens.resize(whisper_tokenize(ctx, params.prompt.c_str(), prompt_tokens.data(), prompt_tokens.size()));
+
+        fprintf(stderr, "\n");
+        fprintf(stderr, "initial prompt: '%s'\n", params.prompt.c_str());
+        fprintf(stderr, "initial tokens: [ ");
+        for (int i = 0; i < (int) prompt_tokens.size(); ++i) {
+            fprintf(stderr, "%d ", prompt_tokens[i]);
+        }
+        fprintf(stderr, "]\n");
+    }
+
+    for (int f = 0; f < (int) params.fname_inp.size(); ++f) {
+        const auto fname_inp = params.fname_inp[f];
+        const auto fname_outp = f < (int)params.fname_outp.size() && !params.fname_outp[f].empty() ? params.fname_outp[f] : params.fname_inp[f];
+
+        std::vector<float> pcmf32; // mono-channel F32 PCM
+        std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
+
+        // WAV input
+        {
+            drwav wav;
+            std::vector<uint8_t> wav_data; // used for pipe input from stdin
+
+            if (fname_inp == "-") {
+                {
+                    uint8_t buf[1024];
+                    while (true)
+                    {
+                        const size_t n = fread(buf, 1, sizeof(buf), stdin);
+                        if (n == 0) {
+                            break;
+                        }
+                        wav_data.insert(wav_data.end(), buf, buf + n);
+                    }
+                }
+
+                if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), nullptr) == false) {
+                    fprintf(stderr, "error: failed to open WAV file from stdin\n");
+                    return 4;
+                }
+
+                fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, wav_data.size());
+            }
+            else if (drwav_init_file(&wav, fname_inp.c_str(), nullptr) == false) {
+                fprintf(stderr, "error: failed to open '%s' as WAV file\n", fname_inp.c_str());
+                return 5;
+            }
+
+           if (wav.channels != 1 && wav.channels != 2) {
+               fprintf(stderr, "error: WAV file '%s' must be mono or stereo\n", fname_inp.c_str());
+               return 6;
+           }
+
+           if (params.diarize && wav.channels != 2 && params.no_timestamps == false) {
+               fprintf(stderr, "error: WAV file '%s' must be stereo for diarization and timestamps have to be enabled\n", fname_inp.c_str());
+               return 6;
+           }
+
+           if (wav.sampleRate != WHISPER_SAMPLE_RATE) {
+               fprintf(stderr, "error: WAV file '%s' must be %i kHz\n", fname_inp.c_str(), WHISPER_SAMPLE_RATE/1000);
+               return 8;
+           }
+
+           if (wav.bitsPerSample != 16) {
+               fprintf(stderr, "error: WAV file '%s' must be 16-bit\n", fname_inp.c_str());
+               return 9;
+           }
+
+            const uint64_t n = wav_data.empty() ? wav.totalPCMFrameCount : wav_data.size()/(wav.channels*wav.bitsPerSample/8);
+
+            std::vector<int16_t> pcm16;
+            pcm16.resize(n*wav.channels);
+            drwav_read_pcm_frames_s16(&wav, n, pcm16.data());
+            drwav_uninit(&wav);
+
+            // convert to mono, float
+            pcmf32.resize(n);
+            if (wav.channels == 1) {
+                for (uint64_t i = 0; i < n; i++) {
+                    pcmf32[i] = float(pcm16[i])/32768.0f;
+                }
+            } else {
+                for (uint64_t i = 0; i < n; i++) {
+                    pcmf32[i] = float(pcm16[2*i] + pcm16[2*i + 1])/65536.0f;
+                }
+            }
+
+            if (params.diarize) {
+                // convert to stereo, float
+                pcmf32s.resize(2);
+
+                pcmf32s[0].resize(n);
+                pcmf32s[1].resize(n);
+                for (uint64_t i = 0; i < n; i++) {
+                    pcmf32s[0][i] = float(pcm16[2*i])/32768.0f;
+                    pcmf32s[1][i] = float(pcm16[2*i + 1])/32768.0f;
+                }
+            }
+        }
+
+        // print system information
+        {
+            fprintf(stderr, "\n");
+            fprintf(stderr, "system_info: n_threads = %d / %d | %s\n",
+                    params.n_threads*params.n_processors, std::thread::hardware_concurrency(), whisper_print_system_info());
+        }
+
+        // print some info about the processing
+        {
+            fprintf(stderr, "\n");
+            if (!whisper_is_multilingual(ctx)) {
+                if (params.language != "en" || params.translate) {
+                    params.language = "en";
+                    params.translate = false;
+                    fprintf(stderr, "%s: WARNING: model is not multilingual, ignoring language and translation options\n", __func__);
+                }
+            }
+            fprintf(stderr, "%s: processing '%s' (%d samples, %.1f sec), %d threads, %d processors, lang = %s, task = %s, timestamps = %d ...\n",
+                    __func__, fname_inp.c_str(), int(pcmf32.size()), float(pcmf32.size())/WHISPER_SAMPLE_RATE,
+                    params.n_threads, params.n_processors,
+                    params.language.c_str(),
+                    params.translate ? "translate" : "transcribe",
+                    params.no_timestamps ? 0 : 1);
+
+            fprintf(stderr, "\n");
+        }
+
+        // run the inference
+        {
+            whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
+
+            wparams.strategy = params.beam_size > 1 ? WHISPER_SAMPLING_BEAM_SEARCH : WHISPER_SAMPLING_GREEDY;
+
+            wparams.print_realtime   = false;
+            wparams.print_progress   = params.print_progress;
+            wparams.print_timestamps = !params.no_timestamps;
+            wparams.print_special    = params.print_special;
+            wparams.translate        = params.translate;
+            wparams.language         = params.language.c_str();
+            wparams.n_threads        = params.n_threads;
+            wparams.n_max_text_ctx   = params.max_context >= 0 ? params.max_context : wparams.n_max_text_ctx;
+            wparams.offset_ms        = params.offset_t_ms;
+            wparams.duration_ms      = params.duration_ms;
+
+            wparams.token_timestamps = params.output_wts || params.max_len > 0;
+            wparams.thold_pt         = params.word_thold;
+            wparams.entropy_thold    = params.entropy_thold;
+            wparams.logprob_thold    = params.logprob_thold;
+            wparams.max_len          = params.output_wts && params.max_len == 0 ? 60 : params.max_len;
+
+            wparams.speed_up         = params.speed_up;
+
+            wparams.greedy.best_of        = params.best_of;
+            wparams.beam_search.beam_size = params.beam_size;
+
+            wparams.prompt_tokens     = prompt_tokens.empty() ? nullptr : prompt_tokens.data();
+            wparams.prompt_n_tokens   = prompt_tokens.empty() ? 0       : prompt_tokens.size();
+
+            whisper_print_user_data user_data = { &params, &pcmf32s };
+
+            // this callback is called on each new segment
+            if (!wparams.print_realtime) {
+                wparams.new_segment_callback           = whisper_print_segment_callback;
+                wparams.new_segment_callback_user_data = &user_data;
+            }
+
+            // example for abort mechanism
+            // in this example, we do not abort the processing, but we could if the flag is set to true
+            // the callback is called before every encoder run - if it returns false, the processing is aborted
+            {
+                static bool is_aborted = false; // NOTE: this should be atomic to avoid data race
+
+                wparams.encoder_begin_callback = [](struct whisper_context * /*ctx*/, void * user_data) {
+                    bool is_aborted = *(bool*)user_data;
+                    return !is_aborted;
+                };
+                wparams.encoder_begin_callback_user_data = &is_aborted;
+            }
+
+            if (whisper_full_parallel(ctx, wparams, pcmf32.data(), pcmf32.size(), params.n_processors) != 0) {
+                fprintf(stderr, "failed to process audio\n");
+                return 10;
+            }
+        }
+    }
+
+    const int n_segments = whisper_full_n_segments(ctx);
+    result.resize(n_segments);
+    for (int i = 0; i < n_segments; ++i) {
+        const char * text = whisper_full_get_segment_text(ctx, i);
+        const int64_t t0 = whisper_full_get_segment_t0(ctx, i);
+        const int64_t t1 = whisper_full_get_segment_t1(ctx, i);
+
+        result[i].emplace_back(to_timestamp(t0, true));
+        result[i].emplace_back(to_timestamp(t1, true));
+        result[i].emplace_back(text);
+    }
+
+    whisper_print_timings(ctx);
+    whisper_free(ctx);
+
+    return 0;
+}
+
+Napi::Object whisper(const Napi::CallbackInfo& info) {
+    Napi::Env env = info.Env();
+    if (info.Length() <= 0 || !info[0].IsObject()) {
+        Napi::TypeError::New(env, "object expected").ThrowAsJavaScriptException();
+    }
+    whisper_params params;
+    std::vector<std::vector<std::string>> result;
+
+    Napi::Object whisper_params = info[0].As<Napi::Object>();
+    std::string language = whisper_params.Get("language").As<Napi::String>();
+    std::string model = whisper_params.Get("model").As<Napi::String>();
+    std::string input = whisper_params.Get("fname_inp").As<Napi::String>();
+
+    params.language = language;
+    params.model = model;
+    params.fname_inp.emplace_back(input);
+
+    // run model
+    run(params, result);
+
+    fprintf(stderr, "RESULT:\n");
+    for (auto sentence:result) {
+        fprintf(stderr, "t0: %s, t1: %s, content: %s \n",
+                sentence[0].c_str(), sentence[1].c_str(), sentence[2].c_str());
+    }
+
+    Napi::Object res = Napi::Array::New(env, result.size());
+    for (u_int32_t i = 0; i < result.size(); ++i) {
+        Napi::Object tmp = Napi::Array::New(env, 3);
+        for (u_int32_t j = 0; j < 3; ++j) {
+            tmp[j] = Napi::String::New(env, result[i][j]);
+        }
+        res[i] = tmp;
+    }
+
+    return res;
+}
+
+
+Napi::Object Init(Napi::Env env, Napi::Object exports) {
+    exports.Set(
+            Napi::String::New(env, "whisper"),
+            Napi::Function::New(env, whisper)
+    );
+    return exports;
+}
+
+NODE_API_MODULE(whisper, Init);

examples/addon.node/index.js

@@ -0,0 +1,27 @@
+const path = require('path');
+const { whisper } = require(path.join(__dirname, '../../build/Release/whisper-addon'));
+
+const whisperParams = {
+    language: 'en',
+    model: path.join(__dirname, '../../models/ggml-base.en.bin'),
+    fname_inp: '',
+};
+
+const arguments = process.argv.slice(2);
+const params = Object.fromEntries(
+    arguments.reduce((pre, item) => {
+        if (item.startsWith("--")) {
+            return [...pre, item.slice(2).split("=")];
+        }
+        return pre;
+    }, []),
+);
+
+for (const key in params) {
+    if (whisperParams.hasOwnProperty(key)) {
+        whisperParams[key] = params[key];
+    }
+}
+
+console.log('whisperParams =', whisperParams);
+console.log(whisper(whisperParams));

examples/addon.node/package.json

@@ -0,0 +1,12 @@
+{
+  "name": "whisper-addon",
+  "version": "0.0.0",
+  "description": "",
+  "main": "index.js",
+  "author": "Qanhe Chen",
+  "license": "MIT",
+  "devDependencies": {
+    "cmake-js": "^7.1.1",
+    "node-addon-api": "^5.0.0"
+  }
+}

examples/bench.wasm/CMakeLists.txt

@@ -8,6 +8,8 @@ add_executable(${TARGET}
     emscripten.cpp
     )
 
+include(DefaultTargetOptions)
+
 target_link_libraries(${TARGET} PRIVATE
     whisper
     )

examples/bench.wasm/emscripten.cpp

@@ -28,6 +28,11 @@ void bench_main(size_t index) {
         return;
     }
 
+    {
+        fprintf(stderr, "\n");
+        fprintf(stderr, "system_info: n_threads = %d / %d | %s\n", n_threads, std::thread::hardware_concurrency(), whisper_print_system_info());
+    }
+
     if (int ret = whisper_encode(ctx, 0, n_threads) != 0) {
         fprintf(stderr, "error: failed to encode model: %d\n", ret);
         return;
@@ -52,7 +57,7 @@ EMSCRIPTEN_BINDINGS(bench) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init(path_model.c_str());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     if (g_worker.joinable()) {
                         g_worker.join();

examples/bench/CMakeLists.txt

@@ -1,3 +1,6 @@
 set(TARGET bench)
 add_executable(${TARGET} bench.cpp)
+
+include(DefaultTargetOptions)
+
 target_link_libraries(${TARGET} PRIVATE whisper ${CMAKE_THREAD_LIBS_INIT})

examples/bench/bench.cpp

@@ -7,6 +7,7 @@
 // command-line parameters
 struct whisper_params {
     int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency());
+    int32_t what = 0; // what to benchmark: 0 - whisper ecoder, 1 - memcpy, 2 - ggml_mul_mat
 
     std::string model = "models/ggml-base.en.bin";
 };
@@ -23,6 +24,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         }
         else if (arg == "-t" || arg == "--threads") { params.n_threads = std::stoi(argv[++i]); }
         else if (arg == "-m" || arg == "--model")   { params.model     = argv[++i]; }
+        else if (arg == "-w" || arg == "--what")    { params.what     = atoi(argv[++i]); }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -33,7 +35,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
     return true;
 }
 
-void whisper_print_usage(int argc, char ** argv, const whisper_params & params) {
+void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params) {
     fprintf(stderr, "\n");
     fprintf(stderr, "usage: %s [options]\n", argv[0]);
     fprintf(stderr, "\n");
@@ -41,19 +43,17 @@ void whisper_print_usage(int argc, char ** argv, const whisper_params & params)
     fprintf(stderr, "  -h,       --help        [default] show this help message and exit\n");
     fprintf(stderr, "  -t N,     --threads N   [%-7d] number of threads to use during computation\n", params.n_threads);
     fprintf(stderr, "  -m FNAME, --model FNAME [%-7s] model path\n",                                  params.model.c_str());
+    fprintf(stderr, "  -w N,     --what N      [%-7d] what to benchmark:\n",                          params.what);
+    fprintf(stderr, "                           %-7s  0 - whisper encoder\n",                         "");
+    fprintf(stderr, "                           %-7s  1 - memcpy\n",                                  "");
+    fprintf(stderr, "                           %-7s  2 - ggml_mul_mat\n",                            "");
     fprintf(stderr, "\n");
 }
 
-int main(int argc, char ** argv) {
-    whisper_params params;
-
-    if (whisper_params_parse(argc, argv, params) == false) {
-        return 1;
-    }
-
+int whisper_bench_encoder(const whisper_params & params) {
     // whisper init
 
-    struct whisper_context * ctx = whisper_init(params.model.c_str());
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     {
         fprintf(stderr, "\n");
@@ -92,3 +92,22 @@ int main(int argc, char ** argv) {
 
     return 0;
 }
+
+int main(int argc, char ** argv) {
+    whisper_params params;
+
+    if (whisper_params_parse(argc, argv, params) == false) {
+        return 1;
+    }
+
+    int ret = -1;
+
+    switch (params.what) {
+        case 0: ret = whisper_bench_encoder(params);                break;
+        case 1: ret = whisper_bench_memcpy(params.n_threads);       break;
+        case 2: ret = whisper_bench_ggml_mul_mat(params.n_threads); break;
+        default: fprintf(stderr, "error: unknown benchmark: %d\n", params.what); break;
+    }
+
+    return ret;
+}

examples/command.wasm/CMakeLists.txt

@@ -8,6 +8,8 @@ add_executable(${TARGET}
     emscripten.cpp
     )
 
+include(DefaultTargetOptions)
+
 target_link_libraries(${TARGET} PRIVATE
     whisper
     )

examples/command.wasm/emscripten.cpp

@@ -324,7 +324,7 @@ EMSCRIPTEN_BINDINGS(command) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init(path_model.c_str());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     g_running = true;
                     if (g_worker.joinable()) {

examples/command/CMakeLists.txt

@@ -2,6 +2,9 @@ if (WHISPER_SUPPORT_SDL2)
     # command
     set(TARGET command)
     add_executable(${TARGET} command.cpp)
+
+    include(DefaultTargetOptions)
+
     target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS})
     target_link_libraries(${TARGET} PRIVATE whisper ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
 endif ()

examples/command/README.md

@@ -9,7 +9,19 @@ More info is available in [issue #171](https://github.com/ggerganov/whisper.cpp/
 
 # On Raspberry Pi, use tiny or base models + "-ac 768" for better performance
 ./command -m ./models/ggml-tiny.en.bin -ac 768 -t 3 -c 0
+```
 
+https://user-images.githubusercontent.com/1991296/204038393-2f846eae-c255-4099-a76d-5735c25c49da.mp4
+
+Web version: [examples/command.wasm](/examples/command.wasm)
+
+## Guided mode
+
+"Guided mode" allows you to specify a list of commands (i.e. strings) and the transcription will be guided to classify your command into one from the list. This can be useful in situations where a device is listening only for a small subset of commands.
+
+Initial tests show that this approach might be extremely efficient in terms of performance, since it integrates very well with the "partial Encoder" idea from #137.
+
+```bash
 # Run in guided mode, the list of allowed commands is in commands.txt
 ./command -m ./models/ggml-base.en.bin -cmd ./examples/command/commands.txt
 
@@ -17,9 +29,8 @@ More info is available in [issue #171](https://github.com/ggerganov/whisper.cpp/
 ./command -m ./models/ggml-tiny.en.bin -cmd ./examples/command/commands.txt -ac 128 -t 3 -c 0
 ```
 
-https://user-images.githubusercontent.com/1991296/204038393-2f846eae-c255-4099-a76d-5735c25c49da.mp4
+https://user-images.githubusercontent.com/1991296/207435352-8fc4ed3f-bde5-4555-9b8b-aeeb76bee969.mp4
 
-Web version: [examples/command.wasm](/examples/command.wasm)
 
 ## Building
 

examples/command/command.cpp

@@ -11,6 +11,7 @@
 #include <SDL.h>
 #include <SDL_audio.h>
 
+#include <sstream>
 #include <cassert>
 #include <cstdio>
 #include <fstream>
@@ -25,7 +26,7 @@
 struct whisper_params {
     int32_t n_threads  = std::min(4, (int32_t) std::thread::hardware_concurrency());
     int32_t prompt_ms  = 5000;
-    int32_t command_ms = 4000;
+    int32_t command_ms = 8000;
     int32_t capture_id = -1;
     int32_t max_tokens = 32;
     int32_t audio_ctx  = 0;
@@ -41,8 +42,9 @@ struct whisper_params {
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
-    std::string fname_out = "";
-    std::string commands  = "";
+    std::string fname_out;
+    std::string commands;
+    std::string prompt;
 };
 
 void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@@ -71,6 +73,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         else if (arg == "-m"   || arg == "--model")         { params.model         = argv[++i]; }
         else if (arg == "-f"   || arg == "--file")          { params.fname_out     = argv[++i]; }
         else if (arg == "-cmd" || arg == "--commands")      { params.commands      = argv[++i]; }
+        else if (arg == "-p"   || arg == "--prompt")        { params.prompt        = argv[++i]; }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -81,7 +84,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
     return true;
 }
 
-void whisper_print_usage(int argc, char ** argv, const whisper_params & params) {
+void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params) {
     fprintf(stderr, "\n");
     fprintf(stderr, "usage: %s [options]\n", argv[0]);
     fprintf(stderr, "\n");
@@ -103,6 +106,7 @@ void whisper_print_usage(int argc, char ** argv, const whisper_params & params)
     fprintf(stderr, "  -m FNAME,   --model FNAME    [%-7s] model path\n",                                  params.model.c_str());
     fprintf(stderr, "  -f FNAME,   --file FNAME     [%-7s] text output file name\n",                       params.fname_out.c_str());
     fprintf(stderr, "  -cmd FNAME, --commands FNAME [%-7s] text file with allowed commands\n",             params.commands.c_str());
+    fprintf(stderr, "  -p,         --prompt         [%-7s] the required activation prompt\n",              params.prompt.c_str());
     fprintf(stderr, "\n");
 }
 
@@ -387,7 +391,7 @@ bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float
     float energy_all  = 0.0f;
     float energy_last = 0.0f;
 
-    for (size_t i = 0; i < n_samples; i++) {
+    for (int i = 0; i < n_samples; i++) {
         energy_all += fabsf(pcmf32[i]);
 
         if (i >= n_samples - n_samples_last) {
@@ -510,6 +514,433 @@ std::vector<std::string> read_allowed_commands(const std::string & fname) {
     return allowed_commands;
 }
 
+std::vector<std::string> get_words(const std::string &txt) {
+    std::vector<std::string> words;
+
+    std::istringstream iss(txt);
+    std::string word;
+    while (iss >> word) {
+        words.push_back(word);
+    }
+
+    return words;
+}
+
+// returns true if no exit event was received
+bool process_sdl_events() {
+    SDL_Event event;
+    while (SDL_PollEvent(&event)) {
+        switch (event.type) {
+            case SDL_QUIT:
+                {
+                    return false;
+                } break;
+            default:
+                break;
+        }
+    }
+
+    return true;
+}
+
+// command-list mode
+// guide the transcription to match the most likely command from a provided list
+int process_command_list(struct whisper_context * ctx, audio_async &audio, const whisper_params &params) {
+    fprintf(stderr, "\n");
+    fprintf(stderr, "%s: guided mode\n", __func__);
+
+    std::vector<std::string> allowed_commands = read_allowed_commands(params.commands);
+
+    if (allowed_commands.empty()) {
+        fprintf(stderr, "%s: error: failed to read allowed commands from '%s'\n", __func__, params.commands.c_str());
+        return 2;
+    }
+
+    int max_len = 0;
+
+    std::vector<std::vector<whisper_token>> allowed_tokens;
+
+    for (const auto & cmd : allowed_commands) {
+        whisper_token tokens[1024];
+        allowed_tokens.emplace_back();
+
+        for (int l = 0; l < (int) cmd.size(); ++l) {
+            // NOTE: very important to add the whitespace !
+            //       the reason is that the first decoded token starts with a whitespace too!
+            std::string ss = std::string(" ") + cmd.substr(0, l + 1);
+
+            const int n = whisper_tokenize(ctx, ss.c_str(), tokens, 1024);
+            if (n < 0) {
+                fprintf(stderr, "%s: error: failed to tokenize command '%s'\n", __func__, cmd.c_str());
+                return 3;
+            }
+
+            if (n == 1) {
+                allowed_tokens.back().push_back(tokens[0]);
+            }
+        }
+
+        max_len = std::max(max_len, (int) cmd.size());
+    }
+
+    fprintf(stderr, "%s: allowed commands [ tokens ]:\n", __func__);
+    fprintf(stderr, "\n");
+    for (int i = 0; i < (int) allowed_commands.size(); ++i) {
+        fprintf(stderr, "  - \033[1m%-*s\033[0m = [", max_len, allowed_commands[i].c_str());
+        for (const auto & token : allowed_tokens[i]) {
+            fprintf(stderr, " %5d", token);
+        }
+        fprintf(stderr, " ]\n");
+    }
+
+    std::string  k_prompt = "select one from the available words: ";
+    for (int i = 0; i < (int) allowed_commands.size(); ++i) {
+        if (i > 0) {
+            k_prompt += ", ";
+        }
+        k_prompt += allowed_commands[i];
+    }
+    k_prompt += ". selected word: ";
+
+    // tokenize prompt
+    std::vector<whisper_token> k_tokens;
+    {
+        k_tokens.resize(1024);
+        const int n = whisper_tokenize(ctx, k_prompt.c_str(), k_tokens.data(), 1024);
+        if (n < 0) {
+            fprintf(stderr, "%s: error: failed to tokenize prompt '%s'\n", __func__, k_prompt.c_str());
+            return 4;
+        }
+        k_tokens.resize(n);
+    }
+
+    fprintf(stderr, "\n");
+    fprintf(stderr, "%s: prompt: '%s'\n", __func__, k_prompt.c_str());
+    fprintf(stderr, "%s: tokens: [", __func__);
+    for (const auto & token : k_tokens) {
+        fprintf(stderr, " %d", token);
+    }
+    fprintf(stderr, " ]\n");
+
+    fprintf(stderr, "\n");
+    fprintf(stderr, "%s: listening for a command ...\n", __func__);
+    fprintf(stderr, "\n");
+
+    bool is_running  = true;
+
+    std::vector<float> pcmf32_cur;
+    std::vector<float> pcmf32_prompt;
+
+    // main loop
+    while (is_running) {
+        // handle Ctrl + C
+        is_running = process_sdl_events();
+
+        // delay
+        std::this_thread::sleep_for(std::chrono::milliseconds(100));
+
+        audio.get(2000, pcmf32_cur);
+
+        if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
+            fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
+
+            const auto t_start = std::chrono::high_resolution_clock::now();
+
+            whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
+
+            wparams.print_progress   = false;
+            wparams.print_special    = params.print_special;
+            wparams.print_realtime   = false;
+            wparams.print_timestamps = !params.no_timestamps;
+            wparams.translate        = params.translate;
+            wparams.no_context       = true;
+            wparams.single_segment   = true;
+            wparams.max_tokens       = 1;
+            wparams.language         = params.language.c_str();
+            wparams.n_threads        = params.n_threads;
+
+            wparams.audio_ctx        = params.audio_ctx;
+            wparams.speed_up         = params.speed_up;
+
+            wparams.prompt_tokens    = k_tokens.data();
+            wparams.prompt_n_tokens  = k_tokens.size();
+
+            // run the transformer and a single decoding pass
+            if (whisper_full(ctx, wparams, pcmf32_cur.data(), pcmf32_cur.size()) != 0) {
+                fprintf(stderr, "%s: ERROR: whisper_full() failed\n", __func__);
+                break;
+            }
+
+            // estimate command probability
+            // NOTE: not optimal
+            {
+                const auto * logits = whisper_get_logits(ctx);
+
+                std::vector<float> probs(whisper_n_vocab(ctx), 0.0f);
+
+                // compute probs from logits via softmax
+                {
+                    float max = -1e9;
+                    for (int i = 0; i < (int) probs.size(); ++i) {
+                        max = std::max(max, logits[i]);
+                    }
+
+                    float sum = 0.0f;
+                    for (int i = 0; i < (int) probs.size(); ++i) {
+                        probs[i] = expf(logits[i] - max);
+                        sum += probs[i];
+                    }
+
+                    for (int i = 0; i < (int) probs.size(); ++i) {
+                        probs[i] /= sum;
+                    }
+                }
+
+                std::vector<std::pair<float, int>> probs_id;
+
+                double psum = 0.0;
+                for (int i = 0; i < (int) allowed_commands.size(); ++i) {
+                    probs_id.emplace_back(probs[allowed_tokens[i][0]], i);
+                    for (int j = 1; j < (int) allowed_tokens[i].size(); ++j) {
+                        probs_id.back().first += probs[allowed_tokens[i][j]];
+                    }
+                    probs_id.back().first /= allowed_tokens[i].size();
+                    psum += probs_id.back().first;
+                }
+
+                // normalize
+                for (auto & p : probs_id) {
+                    p.first /= psum;
+                }
+
+                // sort descending
+                {
+                    using pair_type = decltype(probs_id)::value_type;
+                    std::sort(probs_id.begin(), probs_id.end(), [](const pair_type & a, const pair_type & b) {
+                        return a.first > b.first;
+                    });
+                }
+
+                // print the commands and the respective probabilities
+                {
+                    fprintf(stdout, "\n");
+                    for (const auto & cmd : probs_id) {
+                        fprintf(stdout, "%s: %s%-*s%s = %f | ", __func__, "\033[1m", max_len, allowed_commands[cmd.second].c_str(), "\033[0m", cmd.first);
+                        for (int token : allowed_tokens[cmd.second]) {
+                            fprintf(stdout, "'%4s' %f ", whisper_token_to_str(ctx, token), probs[token]);
+                        }
+                        fprintf(stdout, "\n");
+                    }
+                }
+
+                // best command
+                {
+                    const auto t_end = std::chrono::high_resolution_clock::now();
+
+                    const float prob = probs_id[0].first;
+                    const int index = probs_id[0].second;
+
+                    fprintf(stdout, "\n");
+                    fprintf(stdout, "%s: detected command: %s%s%s | p = %f | t = %d ms\n", __func__,
+                            "\033[1m", allowed_commands[index].c_str(), "\033[0m", prob,
+                            (int) std::chrono::duration_cast<std::chrono::milliseconds>(t_end - t_start).count());
+                    fprintf(stdout, "\n");
+                }
+            }
+
+            audio.clear();
+        }
+    }
+
+    return 0;
+}
+
+// always-prompt mode
+// transcribe the voice into text after valid prompt
+int always_prompt_transcription(struct whisper_context * ctx, audio_async & audio, const whisper_params & params) {
+    bool is_running = true;
+    bool ask_prompt = true;
+
+    float prob = 0.0f;
+
+    std::vector<float> pcmf32_cur;
+
+    const std::string k_prompt = params.prompt;
+
+    const int k_prompt_length = get_words(k_prompt).size();
+
+    fprintf(stderr, "\n");
+    fprintf(stderr, "%s: always-prompt mode\n", __func__);
+
+    // main loop
+    while (is_running) {
+        // handle Ctrl + C
+        is_running = process_sdl_events();
+
+        // delay
+        std::this_thread::sleep_for(std::chrono::milliseconds(100));
+
+        if (ask_prompt) {
+            fprintf(stdout, "\n");
+            fprintf(stdout, "%s: The prompt is: '%s%s%s'\n", __func__, "\033[1m", k_prompt.c_str(), "\033[0m");
+            fprintf(stdout, "\n");
+
+            ask_prompt = false;
+        }
+
+        {
+            audio.get(2000, pcmf32_cur);
+
+            if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
+                fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
+
+                int64_t t_ms = 0;
+
+                // detect the commands
+                audio.get(params.command_ms, pcmf32_cur);
+
+                const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob, t_ms));
+
+                const auto words = get_words(txt);
+
+                std::string prompt;
+                std::string command;
+
+                for (int i = 0; i < (int) words.size(); ++i) {
+                    if (i < k_prompt_length) {
+                        prompt += words[i] + " ";
+                    } else {
+                        command += words[i] + " ";
+                    }
+                }
+
+                const float sim = similarity(prompt, k_prompt);
+
+                //debug
+                //fprintf(stdout, "command size: %i\n", command_length);
+
+                if ((sim > 0.7f) && (command.size() > 0)) {
+                    fprintf(stdout, "%s: Command '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", command.c_str(), "\033[0m", (int) t_ms);
+                }
+
+                fprintf(stdout, "\n");
+
+                audio.clear();
+            }
+        }
+    }
+
+    return 0;
+}
+
+// general-purpose mode
+// freely transcribe the voice into text
+int process_general_transcription(struct whisper_context * ctx, audio_async &audio, const whisper_params &params) {
+    bool is_running  = true;
+    bool have_prompt = false;
+    bool ask_prompt  = true;
+
+    float prob0 = 0.0f;
+    float prob  = 0.0f;
+
+    std::vector<float> pcmf32_cur;
+    std::vector<float> pcmf32_prompt;
+
+    const std::string k_prompt = "Ok Whisper, start listening for commands.";
+
+    fprintf(stderr, "\n");
+    fprintf(stderr, "%s: general-purpose mode\n", __func__);
+
+    // main loop
+    while (is_running) {
+        // handle Ctrl + C
+        is_running = process_sdl_events();
+
+        // delay
+        std::this_thread::sleep_for(std::chrono::milliseconds(100));
+
+        if (ask_prompt) {
+            fprintf(stdout, "\n");
+            fprintf(stdout, "%s: Say the following phrase: '%s%s%s'\n", __func__, "\033[1m", k_prompt.c_str(), "\033[0m");
+            fprintf(stdout, "\n");
+
+            ask_prompt = false;
+        }
+
+        {
+            audio.get(2000, pcmf32_cur);
+
+            if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
+                fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
+
+                int64_t t_ms = 0;
+
+                if (!have_prompt) {
+                    // wait for activation phrase
+                    audio.get(params.prompt_ms, pcmf32_cur);
+
+                    const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob0, t_ms));
+
+                    fprintf(stdout, "%s: Heard '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", txt.c_str(), "\033[0m", (int) t_ms);
+
+                    const float sim = similarity(txt, k_prompt);
+
+                    if (txt.length() < 0.8*k_prompt.length() || txt.length() > 1.2*k_prompt.length() || sim < 0.8f) {
+                        fprintf(stdout, "%s: WARNING: prompt not recognized, try again\n", __func__);
+                        ask_prompt = true;
+                    } else {
+                        fprintf(stdout, "\n");
+                        fprintf(stdout, "%s: The prompt has been recognized!\n", __func__);
+                        fprintf(stdout, "%s: Waiting for voice commands ...\n", __func__);
+                        fprintf(stdout, "\n");
+
+                        // save the audio for the prompt
+                        pcmf32_prompt = pcmf32_cur;
+                        have_prompt = true;
+                    }
+                } else {
+                    // we have heard the activation phrase, now detect the commands
+                    audio.get(params.command_ms, pcmf32_cur);
+
+                    // prepend the prompt audio
+                    pcmf32_cur.insert(pcmf32_cur.begin(), pcmf32_prompt.begin(), pcmf32_prompt.end());
+
+                    const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob, t_ms));
+
+                    prob = 100.0f*(prob - prob0);
+
+                    //fprintf(stdout, "%s: heard '%s'\n", __func__, txt.c_str());
+
+                    // find the prompt in the text
+                    float best_sim = 0.0f;
+                    size_t best_len = 0;
+                    for (int n = 0.8*k_prompt.size(); n <= 1.2*k_prompt.size(); ++n) {
+                        const auto prompt = txt.substr(0, n);
+
+                        const float sim = similarity(prompt, k_prompt);
+
+                        //fprintf(stderr, "%s: prompt = '%s', sim = %f\n", __func__, prompt.c_str(), sim);
+
+                        if (sim > best_sim) {
+                            best_sim = sim;
+                            best_len = n;
+                        }
+                    }
+
+                    const std::string command = ::trim(txt.substr(best_len));
+
+                    fprintf(stdout, "%s: Command '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", command.c_str(), "\033[0m", (int) t_ms);
+                    fprintf(stdout, "\n");
+                }
+
+                audio.clear();
+            }
+        }
+    }
+
+    return 0;
+}
+
 int main(int argc, char ** argv) {
     whisper_params params;
 
@@ -525,7 +956,7 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
-    struct whisper_context * ctx = whisper_init(params.model.c_str());
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     // print some info about the processing
     {
@@ -561,300 +992,14 @@ int main(int argc, char ** argv) {
     std::this_thread::sleep_for(std::chrono::milliseconds(1000));
     audio.clear();
 
-    int max_len = 0;
-
-    bool is_running  = true;
-    bool have_prompt = false;
-    bool ask_prompt  = true;
-
-    float prob0 = 0.0f;
-    float prob  = 0.0f;
-
-    std::vector<float> pcmf32_cur;
-    std::vector<float> pcmf32_prompt;
-
-    std::vector<std::string> allowed_commands;
-    std::vector<std::vector<whisper_token>> allowed_tokens;
-
-    std::string k_prompt = "";
-    std::vector<whisper_token> k_tokens;
-
-    if (params.commands != "") {
-        fprintf(stderr, "\n");
-        fprintf(stderr, "%s: guided mode\n", __func__);
-
-        allowed_commands = read_allowed_commands(params.commands);
-
-        if (allowed_commands.empty()) {
-            fprintf(stderr, "%s: error: failed to read allowed commands from '%s'\n", __func__, params.commands.c_str());
-            return 2;
-        }
-
-        for (const auto & cmd : allowed_commands) {
-            whisper_token tokens[1024];
-            allowed_tokens.emplace_back();
-
-            for (int l = 0; l < cmd.size(); ++l) {
-                // NOTE: very important to add the whitespace !
-                //       the reason is that the first decoded token starts with a whitespace too!
-                std::string ss = std::string(" ") + cmd.substr(0, l + 1);
-
-                const int n = whisper_tokenize(ctx, ss.c_str(), tokens, 1024);
-                if (n < 0) {
-                    fprintf(stderr, "%s: error: failed to tokenize command '%s'\n", __func__, cmd.c_str());
-                    return 3;
-                }
-
-                if (n == 1) {
-                    allowed_tokens.back().push_back(tokens[0]);
-                }
-            }
-
-            max_len = std::max(max_len, (int) cmd.size());
-        }
-
-        fprintf(stderr, "%s: allowed commands [ tokens ]:\n", __func__);
-        fprintf(stderr, "\n");
-        for (int i = 0; i < (int) allowed_commands.size(); ++i) {
-            fprintf(stderr, "  - \033[1m%-*s\033[0m = [", max_len, allowed_commands[i].c_str());
-            for (const auto & token : allowed_tokens[i]) {
-                fprintf(stderr, " %5d", token);
-            }
-            fprintf(stderr, " ]\n");
-        }
-
-        k_prompt = "select one from the available words: ";
-        for (int i = 0; i < (int) allowed_commands.size(); ++i) {
-            if (i > 0) {
-                k_prompt += ", ";
-            }
-            k_prompt += allowed_commands[i];
-        }
-        k_prompt += ". selected word: ";
-
-        // tokenize prompt
-        {
-            k_tokens.resize(1024);
-            const int n = whisper_tokenize(ctx, k_prompt.c_str(), k_tokens.data(), 1024);
-            if (n < 0) {
-                fprintf(stderr, "%s: error: failed to tokenize prompt '%s'\n", __func__, k_prompt.c_str());
-                return 4;
-            }
-            k_tokens.resize(n);
-        }
-
-        fprintf(stderr, "\n");
-        fprintf(stderr, "%s: prompt: '%s'\n", __func__, k_prompt.c_str());
-        fprintf(stderr, "%s: tokens: [", __func__);
-        for (const auto & token : k_tokens) {
-            fprintf(stderr, " %d", token);
-        }
-        fprintf(stderr, " ]\n");
-
-        fprintf(stderr, "\n");
-        fprintf(stderr, "%s: listening for a command ...\n", __func__);
-        fprintf(stderr, "\n");
+    int  ret_val = 0;
 
+    if (!params.commands.empty()) {
+        ret_val = process_command_list(ctx, audio, params);
+    } else if (!params.prompt.empty()) {
+        ret_val = always_prompt_transcription(ctx, audio, params);
     } else {
-        fprintf(stderr, "\n");
-        fprintf(stderr, "%s: general-purpose mode\n", __func__);
-
-        k_prompt = "Ok Whisper, start listening for commands.";
-    }
-
-    // main loop
-    while (is_running) {
-        // handle Ctrl + C
-        {
-            SDL_Event event;
-            while (SDL_PollEvent(&event)) {
-                switch (event.type) {
-                    case SDL_QUIT:
-                        {
-                            is_running = false;
-                        } break;
-                    default:
-                        break;
-                }
-            }
-
-            if (!is_running) {
-                break;
-            }
-        }
-
-        // delay
-        std::this_thread::sleep_for(std::chrono::milliseconds(100));
-
-        if (allowed_commands.empty()) {
-            // general-purpose mode
-            // freely transcribe the voice into text
-
-            if (ask_prompt) {
-                fprintf(stdout, "\n");
-                fprintf(stdout, "%s: Say the following phrase: '%s%s%s'\n", __func__, "\033[1m", k_prompt.c_str(), "\033[0m");
-                fprintf(stdout, "\n");
-
-                ask_prompt = false;
-            }
-
-            {
-                int64_t t_ms = 0;
-
-                audio.get(2000, pcmf32_cur);
-
-                if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
-                    fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
-
-                    if (!have_prompt) {
-                        // wait for activation phrase
-                        audio.get(params.prompt_ms, pcmf32_cur);
-
-                        const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob0, t_ms));
-
-                        fprintf(stdout, "%s: Heard '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", txt.c_str(), "\033[0m", (int) t_ms);
-
-                        const float sim = similarity(txt, k_prompt);
-
-                        if (txt.length() < 0.8*k_prompt.length() || txt.length() > 1.2*k_prompt.length() || sim < 0.8f) {
-                            fprintf(stdout, "%s: WARNING: prompt not recognized, try again\n", __func__);
-                            ask_prompt = true;
-                        } else {
-                            fprintf(stdout, "\n");
-                            fprintf(stdout, "%s: The prompt has been recognized!\n", __func__);
-                            fprintf(stdout, "%s: Waiting for voice commands ...\n", __func__);
-                            fprintf(stdout, "\n");
-
-                            // save the audio for the prompt
-                            pcmf32_prompt = pcmf32_cur;
-                            have_prompt = true;
-                        }
-                    } else {
-                        // we have heard the activation phrase, now detect the commands
-                        audio.get(params.command_ms, pcmf32_cur);
-
-                        // prepend the prompt audio
-                        pcmf32_cur.insert(pcmf32_cur.begin(), pcmf32_prompt.begin(), pcmf32_prompt.end());
-
-                        const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob, t_ms));
-
-                        prob = 100.0f*(prob - prob0);
-
-                        //fprintf(stdout, "%s: heard '%s'\n", __func__, txt.c_str());
-
-                        // find the prompt in the text
-                        float best_sim = 0.0f;
-                        size_t best_len = 0;
-                        for (int n = 0.8*k_prompt.size(); n <= 1.2*k_prompt.size(); ++n) {
-                            const auto prompt = txt.substr(0, n);
-
-                            const float sim = similarity(prompt, k_prompt);
-
-                            //fprintf(stderr, "%s: prompt = '%s', sim = %f\n", __func__, prompt.c_str(), sim);
-
-                            if (sim > best_sim) {
-                                best_sim = sim;
-                                best_len = n;
-                            }
-                        }
-
-                        const std::string command = ::trim(txt.substr(best_len));
-
-                        fprintf(stdout, "%s: Command '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", command.c_str(), "\033[0m", (int) t_ms);
-                        fprintf(stdout, "\n");
-                    }
-
-                    audio.clear();
-                }
-            }
-        } else {
-            // command-list mode
-            // guide the transcription to match the most likely command from a provided list
-
-            audio.get(2000, pcmf32_cur);
-
-            if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
-                fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
-
-                const auto t_start = std::chrono::high_resolution_clock::now();
-
-                whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
-
-                wparams.print_progress   = false;
-                wparams.print_special    = params.print_special;
-                wparams.print_realtime   = false;
-                wparams.print_timestamps = !params.no_timestamps;
-                wparams.translate        = params.translate;
-                wparams.no_context       = true;
-                wparams.single_segment   = true;
-                wparams.max_tokens       = 1;
-                wparams.language         = params.language.c_str();
-                wparams.n_threads        = params.n_threads;
-
-                wparams.audio_ctx        = params.audio_ctx;
-                wparams.speed_up         = params.speed_up;
-
-                wparams.prompt_tokens    = k_tokens.data();
-                wparams.prompt_n_tokens  = k_tokens.size();
-
-                // run the transformer and a single decoding pass
-                if (whisper_full(ctx, wparams, pcmf32_cur.data(), pcmf32_cur.size()) != 0) {
-                    fprintf(stderr, "%s: ERROR: whisper_full() failed\n", __func__);
-                    break;
-                }
-
-                const auto * probs = whisper_get_probs(ctx);
-                std::vector<std::pair<float, int>> probs_id;
-
-                double psum = 0.0;
-                for (int i = 0; i < (int) allowed_commands.size(); ++i) {
-                    probs_id.push_back(std::make_pair(probs[allowed_tokens[i][0]], i));
-                    for (int j = 1; j < (int) allowed_tokens[i].size(); ++j) {
-                        probs_id.back().first += probs[allowed_tokens[i][j]];
-                    }
-                    probs_id.back().first /= allowed_tokens[i].size();
-                    psum += probs_id.back().first;
-                }
-
-                // normalize
-                for (auto & p : probs_id) {
-                    p.first /= psum;
-                }
-
-                // sort descending
-                {
-                    using pair_type = decltype(probs_id)::value_type;
-                    std::sort(probs_id.begin(), probs_id.end(), [](const pair_type & a, const pair_type & b) {
-                        return a.first > b.first;
-                    });
-                }
-
-                // print the commands and the respective probabilities
-                {
-                    fprintf(stdout, "\n");
-                    for (const auto & cmd : probs_id) {
-                        fprintf(stdout, "%s: %s%-*s%s = %f | ", __func__, "\033[1m", max_len, allowed_commands[cmd.second].c_str(), "\033[0m", cmd.first);
-                        for (int i = 0; i < (int) allowed_tokens[cmd.second].size(); ++i) {
-                            fprintf(stdout, "'%4s' %f ", whisper_token_to_str(ctx, allowed_tokens[cmd.second][i]), probs[allowed_tokens[cmd.second][i]]);
-                        }
-                        fprintf(stdout, "\n");
-                    }
-                }
-
-                // best command
-                {
-                    fprintf(stdout, "\n");
-                    fprintf(stdout, "%s: detected command: %s%s%s | p = %f | t = %d ms\n", __func__,
-                            "\033[1m", allowed_commands[probs_id[0].second].c_str(), "\033[0m", probs_id[0].first,
-                            (int) std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::high_resolution_clock::now() - t_start).count());
-                    fprintf(stdout, "\n");
-                }
-
-                const auto t_end = std::chrono::high_resolution_clock::now();
-
-                audio.clear();
-            }
-        }
+        ret_val = process_general_transcription(ctx, audio, params);
     }
 
     audio.pause();
@@ -862,5 +1007,5 @@ int main(int argc, char ** argv) {
     whisper_print_timings(ctx);
     whisper_free(ctx);
 
-    return 0;
+    return ret_val;
 }

examples/helpers.js

@@ -8,7 +8,7 @@ function convertTypedArray(src, type) {
 
 var printTextarea = (function() {
     var element = document.getElementById('output');
-    if (element) element.alue = ''; // clear browser cache
+    if (element) element.value = ''; // clear browser cache
     return function(text) {
         if (arguments.length > 1) text = Array.prototype.slice.call(arguments).join(' ');
         console.log(text);
@@ -88,11 +88,15 @@ async function fetchRemote(url, cbProgress, cbPrint) {
 // - check if the data is already in the IndexedDB
 // - if not, fetch it from the remote URL and store it in the IndexedDB
 function loadRemote(url, dst, size_mb, cbProgress, cbReady, cbCancel, cbPrint) {
-    // query the storage quota and print it
-    navigator.storage.estimate().then(function (estimate) {
-        cbPrint('loadRemote: storage quota: ' + estimate.quota + ' bytes');
-        cbPrint('loadRemote: storage usage: ' + estimate.usage + ' bytes');
-    });
+    if (!navigator.storage || !navigator.storage.estimate) {
+        cbPrint('loadRemote: navigator.storage.estimate() is not supported');
+    } else {
+        // query the storage quota and print it
+        navigator.storage.estimate().then(function (estimate) {
+            cbPrint('loadRemote: storage quota: ' + estimate.quota + ' bytes');
+            cbPrint('loadRemote: storage usage: ' + estimate.usage + ' bytes');
+        });
+    }
 
     // check if the data is already in the IndexedDB
     var rq = indexedDB.open(dbName, dbVersion);

examples/livestream.sh

@@ -100,7 +100,7 @@ while [ $running -eq 1 ]; do
         err=$(cat /tmp/whisper-live.err | wc -l)
     done
 
-    ./main -t 8 -m ./models/ggml-base.en.bin -f /tmp/whisper-live.wav --no-timestamps -otxt 2> /tmp/whispererr | tail -n 1
+    ./main -t 8 -m ./models/ggml-${model}.bin -f /tmp/whisper-live.wav --no-timestamps -otxt 2> /tmp/whispererr | tail -n 1
 
     while [ $SECONDS -lt $((($i+1)*$step_s)) ]; do
         sleep 1

examples/main/CMakeLists.txt

@@ -1,3 +1,6 @@
 set(TARGET main)
 add_executable(${TARGET} main.cpp)
+
+include(DefaultTargetOptions)
+
 target_link_libraries(${TARGET} PRIVATE whisper ${CMAKE_THREAD_LIBS_INIT})

examples/main/README.md

@@ -9,25 +9,35 @@ It can be used as a reference for using the `whisper.cpp` library in other proje
 usage: ./main [options] file0.wav file1.wav ...
 
 options:
-  -h,       --help          [default] show this help message and exit
-  -t N,     --threads N     [4      ] number of threads to use during computation
-  -p N,     --processors N  [1      ] number of processors to use during computation
-  -ot N,    --offset-t N    [0      ] time offset in milliseconds
-  -on N,    --offset-n N    [0      ] segment index offset
-  -d  N,    --duration N    [0      ] duration of audio to process in milliseconds
-  -mc N,    --max-context N [-1     ] maximum number of text context tokens to store
-  -ml N,    --max-len N     [0      ] maximum segment length in characters
-  -wt N,    --word-thold N  [0.01   ] word timestamp probability threshold
-  -su,      --speed-up      [false  ] speed up audio by x2 (reduced accuracy)
-  -tr,      --translate     [false  ] translate from source language to english
-  -otxt,    --output-txt    [false  ] output result in a text file
-  -ovtt,    --output-vtt    [false  ] output result in a vtt file
-  -osrt,    --output-srt    [false  ] output result in a srt file
-  -owts,    --output-words  [false  ] output script for generating karaoke video
-  -ps,      --print-special [false  ] print special tokens
-  -pc,      --print-colors  [false  ] print colors
-  -nt,      --no-timestamps [true   ] do not print timestamps
-  -l LANG,  --language LANG [en     ] spoken language
-  -m FNAME, --model FNAME   [models/ggml-base.en.bin] model path
-  -f FNAME, --file FNAME    [       ] input WAV file path
+  -h,        --help              [default] show this help message and exit
+  -t N,      --threads N         [4      ] number of threads to use during computation
+  -p N,      --processors N      [1      ] number of processors to use during computation
+  -ot N,     --offset-t N        [0      ] time offset in milliseconds
+  -on N,     --offset-n N        [0      ] segment index offset
+  -d  N,     --duration N        [0      ] duration of audio to process in milliseconds
+  -mc N,     --max-context N     [-1     ] maximum number of text context tokens to store
+  -ml N,     --max-len N         [0      ] maximum segment length in characters
+  -bo N,     --best-of N         [5      ] number of best candidates to keep
+  -bs N,     --beam-size N       [-1     ] beam size for beam search
+  -wt N,     --word-thold N      [0.01   ] word timestamp probability threshold
+  -et N,     --entropy-thold N   [2.40   ] entropy threshold for decoder fail
+  -lpt N,    --logprob-thold N   [-1.00  ] log probability threshold for decoder fail
+  -su,       --speed-up          [false  ] speed up audio by x2 (reduced accuracy)
+  -tr,       --translate         [false  ] translate from source language to english
+  -di,       --diarize           [false  ] stereo audio diarization
+  -nf,       --no-fallback       [false  ] do not use temperature fallback while decoding
+  -otxt,     --output-txt        [false  ] output result in a text file
+  -ovtt,     --output-vtt        [false  ] output result in a vtt file
+  -osrt,     --output-srt        [false  ] output result in a srt file
+  -owts,     --output-words      [false  ] output script for generating karaoke video
+  -ocsv,     --output-csv        [false  ] output result in a CSV file
+  -of FNAME, --output-file FNAME [       ] output file path (without file extension)
+  -ps,       --print-special     [false  ] print special tokens
+  -pc,       --print-colors      [false  ] print colors
+  -pp,       --print-progress    [false  ] print progress
+  -nt,       --no-timestamps     [true   ] do not print timestamps
+  -l LANG,   --language LANG     [en     ] spoken language ('auto' for auto-detect)
+             --prompt PROMPT     [       ] initial prompt
+  -m FNAME,  --model FNAME       [models/ggml-base.en.bin] model path
+  -f FNAME,  --file FNAME        [       ] input WAV file path
 ```

examples/main/main.cpp

@@ -53,32 +53,39 @@ void replace_all(std::string & s, const std::string & search, const std::string
 // command-line parameters
 struct whisper_params {
     int32_t n_threads    = std::min(4, (int32_t) std::thread::hardware_concurrency());
-    int32_t n_processors = 1;
-    int32_t offset_t_ms  = 0;
-    int32_t offset_n     = 0;
-    int32_t duration_ms  = 0;
+    int32_t n_processors =  1;
+    int32_t offset_t_ms  =  0;
+    int32_t offset_n     =  0;
+    int32_t duration_ms  =  0;
     int32_t max_context  = -1;
-    int32_t max_len      = 0;
+    int32_t max_len      =  0;
+    int32_t best_of      =  5;
+    int32_t beam_size    = -1;
 
-    float word_thold = 0.01f;
+    float word_thold    =  0.01f;
+    float entropy_thold =  2.40f;
+    float logprob_thold = -1.00f;
 
     bool speed_up       = false;
     bool translate      = false;
     bool diarize        = false;
+    bool no_fallback    = false;
     bool output_txt     = false;
     bool output_vtt     = false;
     bool output_srt     = false;
     bool output_wts     = false;
+    bool output_csv     = false;
     bool print_special  = false;
     bool print_colors   = false;
     bool print_progress = false;
     bool no_timestamps  = false;
 
     std::string language = "en";
-    std::string prompt   = "";
+    std::string prompt;
     std::string model    = "models/ggml-base.en.bin";
 
     std::vector<std::string> fname_inp = {};
+    std::vector<std::string> fname_outp = {};
 };
 
 void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@@ -103,14 +110,21 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         else if (arg == "-d"    || arg == "--duration")       { params.duration_ms    = std::stoi(argv[++i]); }
         else if (arg == "-mc"   || arg == "--max-context")    { params.max_context    = std::stoi(argv[++i]); }
         else if (arg == "-ml"   || arg == "--max-len")        { params.max_len        = std::stoi(argv[++i]); }
+        else if (arg == "-bo"   || arg == "--best-of")        { params.best_of        = std::stoi(argv[++i]); }
+        else if (arg == "-bs"   || arg == "--beam-size")      { params.beam_size      = std::stoi(argv[++i]); }
         else if (arg == "-wt"   || arg == "--word-thold")     { params.word_thold     = std::stof(argv[++i]); }
+        else if (arg == "-et"   || arg == "--entropy-thold")  { params.entropy_thold  = std::stof(argv[++i]); }
+        else if (arg == "-lpt"  || arg == "--logprob-thold")  { params.logprob_thold  = std::stof(argv[++i]); }
         else if (arg == "-su"   || arg == "--speed-up")       { params.speed_up       = true; }
         else if (arg == "-tr"   || arg == "--translate")      { params.translate      = true; }
         else if (arg == "-di"   || arg == "--diarize")        { params.diarize        = true; }
+        else if (arg == "-nf"   || arg == "--no-fallback")    { params.no_fallback    = true; }
         else if (arg == "-otxt" || arg == "--output-txt")     { params.output_txt     = true; }
         else if (arg == "-ovtt" || arg == "--output-vtt")     { params.output_vtt     = true; }
         else if (arg == "-osrt" || arg == "--output-srt")     { params.output_srt     = true; }
         else if (arg == "-owts" || arg == "--output-words")   { params.output_wts     = true; }
+        else if (arg == "-ocsv" || arg == "--output-csv")     { params.output_csv     = true; }
+        else if (arg == "-of"   || arg == "--output-file")    { params.fname_outp.emplace_back(argv[++i]); }
         else if (arg == "-ps"   || arg == "--print-special")  { params.print_special  = true; }
         else if (arg == "-pc"   || arg == "--print-colors")   { params.print_colors   = true; }
         else if (arg == "-pp"   || arg == "--print-progress") { params.print_progress = true; }
@@ -118,7 +132,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         else if (arg == "-l"    || arg == "--language")       { params.language       = argv[++i]; }
         else if (                  arg == "--prompt")         { params.prompt         = argv[++i]; }
         else if (arg == "-m"    || arg == "--model")          { params.model          = argv[++i]; }
-        else if (arg == "-f"    || arg == "--file")           { params.fname_inp.push_back(argv[++i]); }
+        else if (arg == "-f"    || arg == "--file")           { params.fname_inp.emplace_back(argv[++i]); }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -129,35 +143,42 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
     return true;
 }
 
-void whisper_print_usage(int argc, char ** argv, const whisper_params & params) {
+void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params) {
     fprintf(stderr, "\n");
     fprintf(stderr, "usage: %s [options] file0.wav file1.wav ...\n", argv[0]);
     fprintf(stderr, "\n");
     fprintf(stderr, "options:\n");
-    fprintf(stderr, "  -h,       --help           [default] show this help message and exit\n");
-    fprintf(stderr, "  -t N,     --threads N      [%-7d] number of threads to use during computation\n",    params.n_threads);
-    fprintf(stderr, "  -p N,     --processors N   [%-7d] number of processors to use during computation\n", params.n_processors);
-    fprintf(stderr, "  -ot N,    --offset-t N     [%-7d] time offset in milliseconds\n",                    params.offset_t_ms);
-    fprintf(stderr, "  -on N,    --offset-n N     [%-7d] segment index offset\n",                           params.offset_n);
-    fprintf(stderr, "  -d  N,    --duration N     [%-7d] duration of audio to process in milliseconds\n",   params.duration_ms);
-    fprintf(stderr, "  -mc N,    --max-context N  [%-7d] maximum number of text context tokens to store\n", params.max_context);
-    fprintf(stderr, "  -ml N,    --max-len N      [%-7d] maximum segment length in characters\n",           params.max_len);
-    fprintf(stderr, "  -wt N,    --word-thold N   [%-7.2f] word timestamp probability threshold\n",         params.word_thold);
-    fprintf(stderr, "  -su,      --speed-up       [%-7s] speed up audio by x2 (reduced accuracy)\n",        params.speed_up ? "true" : "false");
-    fprintf(stderr, "  -tr,      --translate      [%-7s] translate from source language to english\n",      params.translate ? "true" : "false");
-    fprintf(stderr, "  -di,      --diarize        [%-7s] stereo audio diarization\n",                       params.diarize ? "true" : "false");
-    fprintf(stderr, "  -otxt,    --output-txt     [%-7s] output result in a text file\n",                   params.output_txt ? "true" : "false");
-    fprintf(stderr, "  -ovtt,    --output-vtt     [%-7s] output result in a vtt file\n",                    params.output_vtt ? "true" : "false");
-    fprintf(stderr, "  -osrt,    --output-srt     [%-7s] output result in a srt file\n",                    params.output_srt ? "true" : "false");
-    fprintf(stderr, "  -owts,    --output-words   [%-7s] output script for generating karaoke video\n",     params.output_wts ? "true" : "false");
-    fprintf(stderr, "  -ps,      --print-special  [%-7s] print special tokens\n",                           params.print_special ? "true" : "false");
-    fprintf(stderr, "  -pc,      --print-colors   [%-7s] print colors\n",                                   params.print_colors ? "true" : "false");
-    fprintf(stderr, "  -pp,      --print-progress [%-7s] print progress\n",                                 params.print_progress ? "true" : "false");
-    fprintf(stderr, "  -nt,      --no-timestamps  [%-7s] do not print timestamps\n",                        params.no_timestamps ? "false" : "true");
-    fprintf(stderr, "  -l LANG,  --language LANG  [%-7s] spoken language ('auto' for auto-detect)\n",       params.language.c_str());
-    fprintf(stderr, "            --prompt PROMPT  [%-7s] initial prompt\n",                                 params.prompt.c_str());
-    fprintf(stderr, "  -m FNAME, --model FNAME    [%-7s] model path\n",                                     params.model.c_str());
-    fprintf(stderr, "  -f FNAME, --file FNAME     [%-7s] input WAV file path\n",                            "");
+    fprintf(stderr, "  -h,        --help              [default] show this help message and exit\n");
+    fprintf(stderr, "  -t N,      --threads N         [%-7d] number of threads to use during computation\n",    params.n_threads);
+    fprintf(stderr, "  -p N,      --processors N      [%-7d] number of processors to use during computation\n", params.n_processors);
+    fprintf(stderr, "  -ot N,     --offset-t N        [%-7d] time offset in milliseconds\n",                    params.offset_t_ms);
+    fprintf(stderr, "  -on N,     --offset-n N        [%-7d] segment index offset\n",                           params.offset_n);
+    fprintf(stderr, "  -d  N,     --duration N        [%-7d] duration of audio to process in milliseconds\n",   params.duration_ms);
+    fprintf(stderr, "  -mc N,     --max-context N     [%-7d] maximum number of text context tokens to store\n", params.max_context);
+    fprintf(stderr, "  -ml N,     --max-len N         [%-7d] maximum segment length in characters\n",           params.max_len);
+    fprintf(stderr, "  -bo N,     --best-of N         [%-7d] number of best candidates to keep\n",              params.best_of);
+    fprintf(stderr, "  -bs N,     --beam-size N       [%-7d] beam size for beam search\n",                      params.beam_size);
+    fprintf(stderr, "  -wt N,     --word-thold N      [%-7.2f] word timestamp probability threshold\n",         params.word_thold);
+    fprintf(stderr, "  -et N,     --entropy-thold N   [%-7.2f] entropy threshold for decoder fail\n",           params.entropy_thold);
+    fprintf(stderr, "  -lpt N,    --logprob-thold N   [%-7.2f] log probability threshold for decoder fail\n",   params.logprob_thold);
+    fprintf(stderr, "  -su,       --speed-up          [%-7s] speed up audio by x2 (reduced accuracy)\n",        params.speed_up ? "true" : "false");
+    fprintf(stderr, "  -tr,       --translate         [%-7s] translate from source language to english\n",      params.translate ? "true" : "false");
+    fprintf(stderr, "  -di,       --diarize           [%-7s] stereo audio diarization\n",                       params.diarize ? "true" : "false");
+    fprintf(stderr, "  -nf,       --no-fallback       [%-7s] do not use temperature fallback while decoding\n", params.no_fallback ? "true" : "false");
+    fprintf(stderr, "  -otxt,     --output-txt        [%-7s] output result in a text file\n",                   params.output_txt ? "true" : "false");
+    fprintf(stderr, "  -ovtt,     --output-vtt        [%-7s] output result in a vtt file\n",                    params.output_vtt ? "true" : "false");
+    fprintf(stderr, "  -osrt,     --output-srt        [%-7s] output result in a srt file\n",                    params.output_srt ? "true" : "false");
+    fprintf(stderr, "  -owts,     --output-words      [%-7s] output script for generating karaoke video\n",     params.output_wts ? "true" : "false");
+    fprintf(stderr, "  -ocsv,     --output-csv        [%-7s] output result in a CSV file\n",                    params.output_csv ? "true" : "false");
+    fprintf(stderr, "  -of FNAME, --output-file FNAME [%-7s] output file path (without file extension)\n",      "");
+    fprintf(stderr, "  -ps,       --print-special     [%-7s] print special tokens\n",                           params.print_special ? "true" : "false");
+    fprintf(stderr, "  -pc,       --print-colors      [%-7s] print colors\n",                                   params.print_colors ? "true" : "false");
+    fprintf(stderr, "  -pp,       --print-progress    [%-7s] print progress\n",                                 params.print_progress ? "true" : "false");
+    fprintf(stderr, "  -nt,       --no-timestamps     [%-7s] do not print timestamps\n",                        params.no_timestamps ? "false" : "true");
+    fprintf(stderr, "  -l LANG,   --language LANG     [%-7s] spoken language ('auto' for auto-detect)\n",       params.language.c_str());
+    fprintf(stderr, "             --prompt PROMPT     [%-7s] initial prompt\n",                                 params.prompt.c_str());
+    fprintf(stderr, "  -m FNAME,  --model FNAME       [%-7s] model path\n",                                     params.model.c_str());
+    fprintf(stderr, "  -f FNAME,  --file FNAME        [%-7s] input WAV file path\n",                            "");
     fprintf(stderr, "\n");
 }
 
@@ -173,90 +194,81 @@ void whisper_print_segment_callback(struct whisper_context * ctx, int n_new, voi
 
     const int n_segments = whisper_full_n_segments(ctx);
 
+    std::string speaker = "";
+
+    int64_t t0;
+    int64_t t1;
+
     // print the last n_new segments
     const int s0 = n_segments - n_new;
+
     if (s0 == 0) {
         printf("\n");
     }
 
     for (int i = s0; i < n_segments; i++) {
-        if (params.no_timestamps) {
-            if (params.print_colors) {
-                for (int j = 0; j < whisper_full_n_tokens(ctx, i); ++j) {
-                    if (params.print_special == false) {
-                        const whisper_token id = whisper_full_get_token_id(ctx, i, j);
-                        if (id >= whisper_token_eot(ctx)) {
-                            continue;
-                        }
-                    }
-
-                    const char * text = whisper_full_get_token_text(ctx, i, j);
-                    const float  p    = whisper_full_get_token_p   (ctx, i, j);
-
-                    const int col = std::max(0, std::min((int) k_colors.size(), (int) (std::pow(p, 3)*float(k_colors.size()))));
-
-                    printf("%s%s%s", k_colors[col].c_str(), text, "\033[0m");
-                }
-            } else {
-                const char * text = whisper_full_get_segment_text(ctx, i);
-                printf("%s", text);
-            }
-            fflush(stdout);
-        } else {
-            const int64_t t0 = whisper_full_get_segment_t0(ctx, i);
-            const int64_t t1 = whisper_full_get_segment_t1(ctx, i);
-
-            std::string speaker = "";
-
-            if (params.diarize && pcmf32s.size() == 2) {
-                const int64_t n_samples = pcmf32s[0].size();
-
-                const int64_t is0 = timestamp_to_sample(t0, n_samples);
-                const int64_t is1 = timestamp_to_sample(t1, n_samples);
-
-                double energy0 = 0.0f;
-                double energy1 = 0.0f;
-
-                for (int64_t j = is0; j < is1; j++) {
-                    energy0 += fabs(pcmf32s[0][j]);
-                    energy1 += fabs(pcmf32s[1][j]);
-                }
-
-                if (energy0 > 1.1*energy1) {
-                    speaker = "(speaker 0)";
-                } else if (energy1 > 1.1*energy0) {
-                    speaker = "(speaker 1)";
-                } else {
-                    speaker = "(speaker ?)";
-                }
-
-                //printf("is0 = %lld, is1 = %lld, energy0 = %f, energy1 = %f, %s\n", is0, is1, energy0, energy1, speaker.c_str());
-            }
-
-            if (params.print_colors) {
-                printf("[%s --> %s]  ", to_timestamp(t0).c_str(), to_timestamp(t1).c_str());
-                for (int j = 0; j < whisper_full_n_tokens(ctx, i); ++j) {
-                    if (params.print_special == false) {
-                        const whisper_token id = whisper_full_get_token_id(ctx, i, j);
-                        if (id >= whisper_token_eot(ctx)) {
-                            continue;
-                        }
-                    }
-
-                    const char * text = whisper_full_get_token_text(ctx, i, j);
-                    const float  p    = whisper_full_get_token_p   (ctx, i, j);
-
-                    const int col = std::max(0, std::min((int) k_colors.size(), (int) (std::pow(p, 3)*float(k_colors.size()))));
-
-                    printf("%s%s%s%s", speaker.c_str(), k_colors[col].c_str(), text, "\033[0m");
-                }
-                printf("\n");
-            } else {
-                const char * text = whisper_full_get_segment_text(ctx, i);
-
-                printf("[%s --> %s]  %s%s\n", to_timestamp(t0).c_str(), to_timestamp(t1).c_str(), speaker.c_str(), text);
-            }
+        if (!params.no_timestamps || params.diarize) {
+            t0 = whisper_full_get_segment_t0(ctx, i);
+            t1 = whisper_full_get_segment_t1(ctx, i);
         }
+
+        if (!params.no_timestamps) {
+            printf("[%s --> %s]  ", to_timestamp(t0).c_str(), to_timestamp(t1).c_str());
+        }
+
+        if (params.diarize && pcmf32s.size() == 2) {
+            const int64_t n_samples = pcmf32s[0].size();
+
+            const int64_t is0 = timestamp_to_sample(t0, n_samples);
+            const int64_t is1 = timestamp_to_sample(t1, n_samples);
+
+            double energy0 = 0.0f;
+            double energy1 = 0.0f;
+
+            for (int64_t j = is0; j < is1; j++) {
+                energy0 += fabs(pcmf32s[0][j]);
+                energy1 += fabs(pcmf32s[1][j]);
+            }
+
+            if (energy0 > 1.1*energy1) {
+                speaker = "(speaker 0)";
+            } else if (energy1 > 1.1*energy0) {
+                speaker = "(speaker 1)";
+            } else {
+                speaker = "(speaker ?)";
+            }
+
+            //printf("is0 = %lld, is1 = %lld, energy0 = %f, energy1 = %f, %s\n", is0, is1, energy0, energy1, speaker.c_str());
+        }
+
+        if (params.print_colors) {
+            for (int j = 0; j < whisper_full_n_tokens(ctx, i); ++j) {
+                if (params.print_special == false) {
+                    const whisper_token id = whisper_full_get_token_id(ctx, i, j);
+                    if (id >= whisper_token_eot(ctx)) {
+                        continue;
+                    }
+                }
+
+                const char * text = whisper_full_get_token_text(ctx, i, j);
+                const float  p    = whisper_full_get_token_p   (ctx, i, j);
+
+                const int col = std::max(0, std::min((int) k_colors.size() - 1, (int) (std::pow(p, 3)*float(k_colors.size()))));
+
+                printf("%s%s%s%s", speaker.c_str(), k_colors[col].c_str(), text, "\033[0m");
+            }
+        } else {
+            const char * text = whisper_full_get_segment_text(ctx, i);
+
+            printf("%s%s", speaker.c_str(), text);
+        }
+
+        // with timestamps or speakers: each segment on new line
+        if (!params.no_timestamps || params.diarize) {
+            printf("\n");
+        }
+
+        fflush(stdout);
     }
 }
 
@@ -325,10 +337,32 @@ bool output_srt(struct whisper_context * ctx, const char * fname, const whisper_
     return true;
 }
 
+bool output_csv(struct whisper_context * ctx, const char * fname) {
+    std::ofstream fout(fname);
+    if (!fout.is_open()) {
+        fprintf(stderr, "%s: failed to open '%s' for writing\n", __func__, fname);
+        return false;
+    }
+
+    fprintf(stderr, "%s: saving output to '%s'\n", __func__, fname);
+
+    const int n_segments = whisper_full_n_segments(ctx);
+    for (int i = 0; i < n_segments; ++i) {
+        const char * text = whisper_full_get_segment_text(ctx, i);
+        const int64_t t0 = whisper_full_get_segment_t0(ctx, i);
+        const int64_t t1 = whisper_full_get_segment_t1(ctx, i);
+
+        //need to multiply times returned from whisper_full_get_segment_t{0,1}() by 10 to get milliseconds.
+        fout << 10 * t0 << ", " << 10 * t1 << ", \"" << text    << "\"\n";
+    }
+
+    return true;
+}
+
 // karaoke video generation
 // outputs a bash script that uses ffmpeg to generate a video with the subtitles
 // TODO: font parameter adjustments
-bool output_wts(struct whisper_context * ctx, const char * fname, const char * fname_inp, const whisper_params & params, float t_sec) {
+bool output_wts(struct whisper_context * ctx, const char * fname, const char * fname_inp, const whisper_params & /*params*/, float t_sec) {
     std::ofstream fout(fname);
 
     fprintf(stderr, "%s: saving output to '%s'\n", __func__, fname);
@@ -377,7 +411,6 @@ bool output_wts(struct whisper_context * ctx, const char * fname, const char * f
             txt_ul = "\\ \\ ";
 
             {
-                int ncnt = 0;
                 for (int k = 0; k < n; ++k) {
                     const auto & token2 = tokens[k];
 
@@ -401,8 +434,6 @@ bool output_wts(struct whisper_context * ctx, const char * fname, const char * f
                             txt_ul += "\\ ";
                         }
                     }
-
-                    ncnt += txt.size();
                 }
 
                 ::replace_all(txt_bg, "'", "\u2019");
@@ -461,7 +492,7 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
-    struct whisper_context * ctx = whisper_init(params.model.c_str());
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     if (ctx == nullptr) {
         fprintf(stderr, "error: failed to initialize whisper context\n");
@@ -471,7 +502,7 @@ int main(int argc, char ** argv) {
     // initial prompt
     std::vector<whisper_token> prompt_tokens;
 
-    if (params.prompt.size() > 0) {
+    if (!params.prompt.empty()) {
         prompt_tokens.resize(1024);
         prompt_tokens.resize(whisper_tokenize(ctx, params.prompt.c_str(), prompt_tokens.data(), prompt_tokens.size()));
 
@@ -486,6 +517,7 @@ int main(int argc, char ** argv) {
 
     for (int f = 0; f < (int) params.fname_inp.size(); ++f) {
         const auto fname_inp = params.fname_inp[f];
+		const auto fname_outp = f < (int) params.fname_outp.size() && !params.fname_outp[f].empty() ? params.fname_outp[f] : params.fname_inp[f];
 
         std::vector<float> pcmf32; // mono-channel F32 PCM
         std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
@@ -508,14 +540,14 @@ int main(int argc, char ** argv) {
                     }
                 }
 
-                if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), NULL) == false) {
+                if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), nullptr) == false) {
                     fprintf(stderr, "error: failed to open WAV file from stdin\n");
                     return 4;
                 }
 
                 fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, wav_data.size());
             }
-            else if (drwav_init_file(&wav, fname_inp.c_str(), NULL) == false) {
+            else if (drwav_init_file(&wav, fname_inp.c_str(), nullptr) == false) {
                 fprintf(stderr, "error: failed to open '%s' as WAV file\n", fname_inp.c_str());
                 return 5;
             }
@@ -531,7 +563,7 @@ int main(int argc, char ** argv) {
             }
 
             if (wav.sampleRate != WHISPER_SAMPLE_RATE) {
-                fprintf(stderr, "%s: WAV file '%s' must be 16 kHz\n", argv[0], fname_inp.c_str());
+                fprintf(stderr, "%s: WAV file '%s' must be %i kHz\n", argv[0], fname_inp.c_str(), WHISPER_SAMPLE_RATE/1000);
                 return 8;
             }
 
@@ -550,11 +582,11 @@ int main(int argc, char ** argv) {
             // convert to mono, float
             pcmf32.resize(n);
             if (wav.channels == 1) {
-                for (int i = 0; i < n; i++) {
+                for (uint64_t i = 0; i < n; i++) {
                     pcmf32[i] = float(pcm16[i])/32768.0f;
                 }
             } else {
-                for (int i = 0; i < n; i++) {
+                for (uint64_t i = 0; i < n; i++) {
                     pcmf32[i] = float(pcm16[2*i] + pcm16[2*i + 1])/65536.0f;
                 }
             }
@@ -565,7 +597,7 @@ int main(int argc, char ** argv) {
 
                 pcmf32s[0].resize(n);
                 pcmf32s[1].resize(n);
-                for (int i = 0; i < n; i++) {
+                for (uint64_t i = 0; i < n; i++) {
                     pcmf32s[0][i] = float(pcm16[2*i])/32768.0f;
                     pcmf32s[1][i] = float(pcm16[2*i + 1])/32768.0f;
                 }
@@ -603,6 +635,8 @@ int main(int argc, char ** argv) {
         {
             whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
 
+            wparams.strategy = params.beam_size > 1 ? WHISPER_SAMPLING_BEAM_SEARCH : WHISPER_SAMPLING_GREEDY;
+
             wparams.print_realtime   = false;
             wparams.print_progress   = params.print_progress;
             wparams.print_timestamps = !params.no_timestamps;
@@ -620,8 +654,15 @@ int main(int argc, char ** argv) {
 
             wparams.speed_up         = params.speed_up;
 
-            wparams.prompt_tokens    = prompt_tokens.size() == 0 ? nullptr : prompt_tokens.data();
-            wparams.prompt_n_tokens  = prompt_tokens.size() == 0 ? 0       : prompt_tokens.size();
+            wparams.prompt_tokens     = prompt_tokens.empty() ? nullptr : prompt_tokens.data();
+            wparams.prompt_n_tokens   = prompt_tokens.empty() ? 0       : prompt_tokens.size();
+
+            wparams.greedy.best_of        = params.best_of;
+            wparams.beam_search.beam_size = params.beam_size;
+
+            wparams.temperature_inc  = params.no_fallback ? 0.0f : wparams.temperature_inc;
+            wparams.entropy_thold    = params.entropy_thold;
+            wparams.logprob_thold    = params.logprob_thold;
 
             whisper_print_user_data user_data = { &params, &pcmf32s };
 
@@ -637,7 +678,7 @@ int main(int argc, char ** argv) {
             {
                 static bool is_aborted = false; // NOTE: this should be atomic to avoid data race
 
-                wparams.encoder_begin_callback = [](struct whisper_context * ctx, void * user_data) {
+                wparams.encoder_begin_callback = [](struct whisper_context * /*ctx*/, void * user_data) {
                     bool is_aborted = *(bool*)user_data;
                     return !is_aborted;
                 };
@@ -656,27 +697,34 @@ int main(int argc, char ** argv) {
 
             // output to text file
             if (params.output_txt) {
-                const auto fname_txt = fname_inp + ".txt";
+                const auto fname_txt = fname_outp + ".txt";
                 output_txt(ctx, fname_txt.c_str());
             }
 
             // output to VTT file
             if (params.output_vtt) {
-                const auto fname_vtt = fname_inp + ".vtt";
+                const auto fname_vtt = fname_outp + ".vtt";
                 output_vtt(ctx, fname_vtt.c_str());
             }
 
             // output to SRT file
             if (params.output_srt) {
-                const auto fname_srt = fname_inp + ".srt";
+                const auto fname_srt = fname_outp + ".srt";
                 output_srt(ctx, fname_srt.c_str(), params);
             }
 
             // output to WTS file
             if (params.output_wts) {
-                const auto fname_wts = fname_inp + ".wts";
+                const auto fname_wts = fname_outp + ".wts";
                 output_wts(ctx, fname_wts.c_str(), fname_inp.c_str(), params, float(pcmf32.size() + 1000)/WHISPER_SAMPLE_RATE);
             }
+
+	    // output to CSV file
+            if (params.output_csv) {
+                const auto fname_csv = fname_outp + ".csv";
+                output_csv(ctx, fname_csv.c_str());
+            }
+
         }
     }
 

examples/stream.wasm/CMakeLists.txt

@@ -8,6 +8,8 @@ add_executable(${TARGET}
     emscripten.cpp
     )
 
+include(DefaultTargetOptions)
+
 target_link_libraries(${TARGET} PRIVATE
     whisper
     )

examples/stream.wasm/emscripten.cpp

@@ -49,6 +49,9 @@ void stream_main(size_t index) {
     wparams.max_tokens       = 32;
     wparams.audio_ctx        = 768; // partial encoder context for better performance
 
+    // disable temperature fallback
+    wparams.temperature_inc  = -1.0f;
+
     wparams.language         = "en";
 
     printf("stream: using %d threads\n", wparams.n_threads);
@@ -129,7 +132,7 @@ EMSCRIPTEN_BINDINGS(stream) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init(path_model.c_str());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     g_running = true;
                     if (g_worker.joinable()) {

examples/stream/CMakeLists.txt

@@ -2,6 +2,9 @@ if (WHISPER_SUPPORT_SDL2)
     # stream
     set(TARGET stream)
     add_executable(${TARGET} stream.cpp)
+
+    include(DefaultTargetOptions)
+
     target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS})
     target_link_libraries(${TARGET} PRIVATE whisper ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
 endif ()

examples/stream/stream.cpp

@@ -8,6 +8,7 @@
 #include <SDL.h>
 #include <SDL_audio.h>
 
+#include <atomic>
 #include <cassert>
 #include <cstdio>
 #include <string>
@@ -51,7 +52,7 @@ struct whisper_params {
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
-    std::string fname_out = "";
+    std::string fname_out;
 };
 
 void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@@ -90,7 +91,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
     return true;
 }
 
-void whisper_print_usage(int argc, char ** argv, const whisper_params & params) {
+void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params) {
     fprintf(stderr, "\n");
     fprintf(stderr, "usage: %s [options]\n", argv[0]);
     fprintf(stderr, "\n");
@@ -144,8 +145,8 @@ private:
     int m_len_ms = 0;
     int m_sample_rate = 0;
 
-    bool       m_running = false;
-    std::mutex m_mutex;
+    std::atomic_bool m_running;
+    std::mutex       m_mutex;
 
     std::vector<float> m_audio;
     std::vector<float> m_audio_new;
@@ -155,6 +156,8 @@ private:
 
 audio_async::audio_async(int len_ms) {
     m_len_ms = len_ms;
+
+    m_running = false;
 }
 
 audio_async::~audio_async() {
@@ -391,7 +394,7 @@ bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float
     float energy_all  = 0.0f;
     float energy_last = 0.0f;
 
-    for (size_t i = 0; i < n_samples; i++) {
+    for (int i = 0; i < n_samples; i++) {
         energy_all += fabsf(pcmf32[i]);
 
         if (i >= n_samples - n_samples_last) {
@@ -420,20 +423,21 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
-    params.keep_ms = std::min(params.keep_ms, params.step_ms); // cannot be more than step_ms
+    params.keep_ms   = std::min(params.keep_ms,   params.step_ms);
+    params.length_ms = std::max(params.length_ms, params.step_ms);
 
     const int n_samples_step = (params.step_ms  *1e-3)*WHISPER_SAMPLE_RATE;
     const int n_samples_len  = (params.length_ms*1e-3)*WHISPER_SAMPLE_RATE;
     const int n_samples_keep = (params.keep_ms  *1e-3)*WHISPER_SAMPLE_RATE;
     const int n_samples_30s  = (30000           *1e-3)*WHISPER_SAMPLE_RATE;
 
-    const int n_new_line = params.length_ms / params.step_ms - 1; // number of steps to print new line
-
     const bool use_vad = n_samples_step <= 0; // sliding window mode uses VAD
 
-    params.no_timestamps = !use_vad;
-    params.no_context    = use_vad;
-    params.max_tokens    = 0;
+    const int n_new_line = !use_vad ? std::max(1, params.length_ms / params.step_ms - 1) : 1; // number of steps to print new line
+
+    params.no_timestamps  = !use_vad;
+    params.no_context    |= use_vad;
+    params.max_tokens     = 0;
 
     // init audio
 
@@ -453,10 +457,10 @@ int main(int argc, char ** argv) {
         exit(0);
     }
 
-    struct whisper_context * ctx = whisper_init(params.model.c_str());
+    struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
 
     std::vector<float> pcmf32    (n_samples_30s, 0.0f);
-    std::vector<float> pcmf32_old(n_samples_30s, 0.0f);
+    std::vector<float> pcmf32_old;
     std::vector<float> pcmf32_new(n_samples_30s, 0.0f);
 
     std::vector<whisper_token> prompt_tokens;
@@ -483,7 +487,7 @@ int main(int argc, char ** argv) {
                 params.no_timestamps ? 0 : 1);
 
         if (!use_vad) {
-            fprintf(stderr, "%s: n_new_line = %d\n", __func__, n_new_line);
+            fprintf(stderr, "%s: n_new_line = %d, no_context = %d\n", __func__, n_new_line, params.no_context);
         } else {
             fprintf(stderr, "%s: using VAD, will transcribe on speech activity\n", __func__);
         }
@@ -612,6 +616,9 @@ int main(int argc, char ** argv) {
             wparams.audio_ctx        = params.audio_ctx;
             wparams.speed_up         = params.speed_up;
 
+            // disable temperature fallback
+            wparams.temperature_inc  = -1.0f;
+
             wparams.prompt_tokens    = params.no_context ? nullptr : prompt_tokens.data();
             wparams.prompt_n_tokens  = params.no_context ? 0       : prompt_tokens.size();
 

examples/talk.wasm/CMakeLists.txt

@@ -9,6 +9,8 @@ add_executable(${TARGET}
     gpt-2.cpp
     )
 
+include(DefaultTargetOptions)
+
 target_link_libraries(${TARGET} PRIVATE
     whisper
     )
@@ -31,8 +33,8 @@ set_target_properties(${TARGET} PROPERTIES LINK_FLAGS " \
     --bind \
     -s USE_PTHREADS=1 \
     -s PTHREAD_POOL_SIZE=8 \
-    -s INITIAL_MEMORY=1600MB \
-    -s TOTAL_MEMORY=1600MB \
+    -s INITIAL_MEMORY=1800MB \
+    -s TOTAL_MEMORY=1800MB \
     -s FORCE_FILESYSTEM=1 \
     -s EXPORTED_RUNTIME_METHODS=\"['print', 'printErr', 'ccall', 'cwrap']\" \
     ${EXTRA_FLAGS} \

examples/talk.wasm/README.md

@@ -36,7 +36,7 @@ In order to run this demo efficiently, you need to have the following:
 - Latest Chrome or Firefox browser (Safari is not supported)
 - Run this on a desktop or laptop with modern CPU (a mobile phone will likely not be good enough)
 - Speak phrases that are no longer than 10 seconds - this is the audio context of the AI
-- The web-page uses about 1.6GB of RAM
+- The web-page uses about 1.8GB of RAM
 
 Notice that this demo is using the smallest GPT-2 model, so the generated text responses are not always very good.
 Also, the prompting strategy can likely be improved to achieve better results.

examples/talk.wasm/emscripten.cpp

@@ -271,7 +271,7 @@ EMSCRIPTEN_BINDINGS(talk) {
     emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init(path_model.c_str());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     g_running = true;
                     if (g_worker.joinable()) {

examples/talk/CMakeLists.txt

@@ -8,6 +8,9 @@ if (WHISPER_SUPPORT_SDL2)
     # TODO: this is temporary
     #       need to export ggml symbols for MSVC, but too lazy ..
     add_executable(${TARGET} talk.cpp gpt-2.cpp ../../ggml.c ../../whisper.cpp)
+
+    include(DefaultTargetOptions)
+
     target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS} ../../)
     target_link_libraries(${TARGET} PRIVATE ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
 endif ()

examples/talk/gpt-2.cpp

@@ -40,7 +40,7 @@ std::vector<gpt_vocab::id> gpt_tokenize(const gpt_vocab & vocab, const std::stri
     // find the longest tokens that form the words:
     std::vector<gpt_vocab::id> tokens;
     for (const auto & word : words) {
-        if (word.size() == 0) continue;
+        if (word.empty()) continue;
 
         int i = 0;
         int n = word.size();
@@ -78,7 +78,7 @@ gpt_vocab::id gpt_sample_top_k_top_p(
         const float * logits,
         int    top_k,
         double top_p,
-        double temp,
+        double /*temp*/,
         std::mt19937 & rng) {
     int n_logits = vocab.id_to_token.size();
 
@@ -86,7 +86,7 @@ gpt_vocab::id gpt_sample_top_k_top_p(
     logits_id.reserve(n_logits);
 
     for (int i = 0; i < n_logits; i++) {
-        logits_id.push_back(std::make_pair(logits[i], i));
+        logits_id.emplace_back(logits[i], i);
     }
 
     // find the top K tokens
@@ -268,7 +268,7 @@ bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab &
             fin.read((char *) &len, sizeof(len));
 
             word.resize(len);
-            fin.read((char *) word.data(), len);
+            fin.read((char *) &word[0], len);
 
             vocab.token_to_id[word] = i;
             vocab.id_to_token[i] = word;
@@ -327,7 +327,7 @@ bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab &
     {
         struct ggml_init_params params;
         params.mem_size   = ctx_size;
-        params.mem_buffer = NULL;
+        params.mem_buffer = nullptr;
 
         model.ctx = ggml_init(params);
         if (!model.ctx) {
@@ -448,7 +448,7 @@ bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab &
             std::string name(length, 0);
             fin.read(&name[0], length);
 
-            if (model.tensors.find(name.data()) == model.tensors.end()) {
+            if (model.tensors.find(name) == model.tensors.end()) {
                 fprintf(stderr, "%s: unknown tensor '%s' in model file\n", __func__, name.data());
                 return false;
             }
@@ -833,7 +833,7 @@ Me too.
 struct gpt2_context * gpt2_init(const char * path_model) {
     gpt2_context * ctx = new gpt2_context;
 
-    ctx->rng = std::mt19937(time(NULL));
+    ctx->rng = std::mt19937(time(nullptr));
 
     // load the model
     {
@@ -841,6 +841,7 @@ struct gpt2_context * gpt2_init(const char * path_model) {
 
         if (!gpt2_model_load(path_model, ctx->model, ctx->vocab)) {
             fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, path_model);
+            delete ctx;
             return nullptr;
         }
 
@@ -884,9 +885,9 @@ std::string gpt2_gen_text(gpt2_context * ctx, const char * text, int max_tokens)
 
     std::string result;
 
-    for (int i = embd.size(); i < embd_inp.size() + n_predict; i++) {
+    for (int i = embd.size(); i < (int) embd_inp.size() + n_predict; i++) {
         // predict
-        if (embd.size() > 0) {
+        if (!embd.empty()) {
             if (!gpt2_eval(ctx->model, ctx->n_threads, n_past, embd, embd_w, mem_per_token)) {
                 printf("gpt-2: failed to generate text\n");
                 return "";

examples/talk/talk.cpp

@@ -39,7 +39,7 @@ struct whisper_params {
     std::string model_wsp = "models/ggml-base.en.bin";
     std::string model_gpt = "models/ggml-gpt-2-117M.bin";
     std::string speak     = "./examples/talk/speak.sh";
-    std::string fname_out = "";
+    std::string fname_out;
 };
 
 void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@@ -79,7 +79,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
     return true;
 }
 
-void whisper_print_usage(int argc, char ** argv, const whisper_params & params) {
+void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params) {
     fprintf(stderr, "\n");
     fprintf(stderr, "usage: %s [options]\n", argv[0]);
     fprintf(stderr, "\n");
@@ -397,7 +397,7 @@ bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float
     float energy_all  = 0.0f;
     float energy_last = 0.0f;
 
-    for (size_t i = 0; i < n_samples; i++) {
+    for (int i = 0; i < n_samples; i++) {
         energy_all += fabsf(pcmf32[i]);
 
         if (i >= n_samples - n_samples_last) {
@@ -498,7 +498,7 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
-    struct whisper_context * ctx_wsp = whisper_init(params.model_wsp.c_str());
+    struct whisper_context * ctx_wsp = whisper_init_from_file(params.model_wsp.c_str());
 
     // gpt init
 
@@ -541,7 +541,6 @@ int main(int argc, char ** argv) {
     bool force_speak = false;
 
     float prob0 = 0.0f;
-    float prob  = 0.0f;
 
     std::vector<float> pcmf32_cur;
     std::vector<float> pcmf32_prompt;
@@ -589,7 +588,7 @@ int main(int argc, char ** argv) {
 
                 audio.get(params.voice_ms, pcmf32_cur);
 
-                std::string text_heard = "";
+                std::string text_heard;
 
                 if (!force_speak) {
                     text_heard = ::trim(::transcribe(ctx_wsp, params, pcmf32_cur, prob0, t_ms));
@@ -611,7 +610,7 @@ int main(int argc, char ** argv) {
                 text_heard = std::regex_replace(text_heard, std::regex("[^a-zA-Z0-9\\.,\\?!\\s\\:\\'\\-]"), "");
 
                 // take first line
-                text_heard = text_heard.substr(0, text_heard.find_first_of("\n"));
+                text_heard = text_heard.substr(0, text_heard.find_first_of('\n'));
 
                 // remove leading and trailing whitespace
                 text_heard = std::regex_replace(text_heard, std::regex("^\\s+"), "");
@@ -641,18 +640,18 @@ int main(int argc, char ** argv) {
 
                     text_to_speak = gpt2_gen_text(ctx_gpt, prompt.c_str(), params.max_tokens);
                     text_to_speak = std::regex_replace(text_to_speak, std::regex("[^a-zA-Z0-9\\.,\\?!\\s\\:\\'\\-]"), "");
-                    text_to_speak = text_to_speak.substr(0, text_to_speak.find_first_of("\n"));
+                    text_to_speak = text_to_speak.substr(0, text_to_speak.find_first_of('\n'));
 
                     // remove first 2 lines of base prompt
                     if (n_iter > 4) {
                         {
-                            const size_t pos = prompt_base.find_first_of("\n");
+                            const size_t pos = prompt_base.find_first_of('\n');
                             if (pos != std::string::npos) {
                                 prompt_base = prompt_base.substr(pos + 1);
                             }
                         }
                         {
-                            const size_t pos = prompt_base.find_first_of("\n");
+                            const size_t pos = prompt_base.find_first_of('\n');
                             if (pos != std::string::npos) {
                                 prompt_base = prompt_base.substr(pos + 1);
                             }

examples/whisper.android/.idea/gradle.xml

@@ -1,5 +1,6 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <project version="4">
+  <component name="GradleMigrationSettings" migrationVersion="1" />
   <component name="GradleSettings">
     <option name="linkedExternalProjectsSettings">
       <GradleProjectSettings>

examples/whisper.android/app/build.gradle

@@ -14,10 +14,6 @@ android {
         versionCode 1
         versionName "1.0"
 
-        ndk {
-            abiFilters 'arm64-v8a', 'x86_64'
-        }
-
         testInstrumentationRunner "androidx.test.runner.AndroidJUnitRunner"
         vectorDrawables {
             useSupportLibrary true
@@ -44,7 +40,7 @@ android {
     composeOptions {
         kotlinCompilerExtensionVersion '1.3.1'
     }
-    ndkVersion "25.0.8528842"
+    ndkVersion "25.1.8937393"
     externalNativeBuild {
         ndkBuild {
             path 'src/main/jni/whisper/Android.mk'

examples/whisper.android/app/src/main/java/com/whispercppdemo/ui/main/MainScreenViewModel.kt

@@ -64,16 +64,21 @@ class MainScreenViewModel(private val application: Application) : ViewModel() {
     private suspend fun copyAssets() = withContext(Dispatchers.IO) {
         modelsPath.mkdirs()
         samplesPath.mkdirs()
-        application.copyData("models", modelsPath, ::printMessage)
+        //application.copyData("models", modelsPath, ::printMessage)
         application.copyData("samples", samplesPath, ::printMessage)
         printMessage("All data copied to working directory.\n")
     }
 
     private suspend fun loadBaseModel() = withContext(Dispatchers.IO) {
         printMessage("Loading model...\n")
-        val firstModel = modelsPath.listFiles()!!.first()
-        whisperContext = WhisperContext.createContext(firstModel.absolutePath)
-        printMessage("Loaded model ${firstModel.name}.\n")
+        val models = application.assets.list("models/")
+        if (models != null) {
+            whisperContext = WhisperContext.createContextFromAsset(application.assets, "models/" + models[0])
+            printMessage("Loaded model ${models[0]}.\n")
+        }
+
+        //val firstModel = modelsPath.listFiles()!!.first()
+        //whisperContext = WhisperContext.createContextFromFile(firstModel.absolutePath)
     }
 
     fun transcribeSample() = viewModelScope.launch {

examples/whisper.android/app/src/main/java/com/whispercppdemo/whisper/LibWhisper.kt

@@ -1,8 +1,15 @@
 package com.whispercppdemo.whisper
 
+import android.content.res.AssetManager
+import android.os.Build
+import android.util.Log
 import kotlinx.coroutines.*
+import java.io.File
+import java.io.InputStream
 import java.util.concurrent.Executors
 
+private const val LOG_TAG = "LibWhisper"
+
 class WhisperContext private constructor(private var ptr: Long) {
     // Meet Whisper C++ constraint: Don't access from more than one thread at a time.
     private val scope: CoroutineScope = CoroutineScope(
@@ -34,23 +41,63 @@ class WhisperContext private constructor(private var ptr: Long) {
     }
 
     companion object {
-        fun createContext(filePath: String): WhisperContext {
+        fun createContextFromFile(filePath: String): WhisperContext {
             val ptr = WhisperLib.initContext(filePath)
             if (ptr == 0L) {
                 throw java.lang.RuntimeException("Couldn't create context with path $filePath")
             }
             return WhisperContext(ptr)
         }
+
+        fun createContextFromInputStream(stream: InputStream): WhisperContext {
+            val ptr = WhisperLib.initContextFromInputStream(stream)
+
+            if (ptr == 0L) {
+                throw java.lang.RuntimeException("Couldn't create context from input stream")
+            }
+            return WhisperContext(ptr)
+        }
+
+        fun createContextFromAsset(assetManager: AssetManager, assetPath: String): WhisperContext {
+            val ptr = WhisperLib.initContextFromAsset(assetManager, assetPath)
+
+            if (ptr == 0L) {
+                throw java.lang.RuntimeException("Couldn't create context from asset $assetPath")
+            }
+            return WhisperContext(ptr)
+        }
     }
 }
 
 private class WhisperLib {
     companion object {
         init {
-            System.loadLibrary("whisper")
+            Log.d(LOG_TAG, "Primary ABI: ${Build.SUPPORTED_ABIS[0]}")
+            var loadVfpv4 = false
+            if (isArmEabiV7a()) {
+                // armeabi-v7a needs runtime detection support
+                val cpuInfo = cpuInfo()
+                cpuInfo?.let {
+                    Log.d(LOG_TAG, "CPU info: $cpuInfo")
+                    if (cpuInfo.contains("vfpv4")) {
+                        Log.d(LOG_TAG, "CPU supports vfpv4")
+                        loadVfpv4 = true
+                    }
+                }
+            }
+
+            if (loadVfpv4) {
+                Log.d(LOG_TAG, "Loading libwhisper_vfpv4.so")
+                System.loadLibrary("whisper_vfpv4")
+            } else {
+                Log.d(LOG_TAG, "Loading libwhisper.so")
+                System.loadLibrary("whisper")
+            }
         }
 
         // JNI methods
+        external fun initContextFromInputStream(inputStream: InputStream): Long
+        external fun initContextFromAsset(assetManager: AssetManager, assetPath: String): Long
         external fun initContext(modelPath: String): Long
         external fun freeContext(contextPtr: Long)
         external fun fullTranscribe(contextPtr: Long, audioData: FloatArray)
@@ -59,3 +106,17 @@ private class WhisperLib {
     }
 }
 
+private fun isArmEabiV7a(): Boolean {
+    return Build.SUPPORTED_ABIS[0].equals("armeabi-v7a")
+}
+
+private fun cpuInfo(): String? {
+    return try {
+        File("/proc/cpuinfo").inputStream().bufferedReader().use {
+            it.readText()
+        }
+    } catch (e: Exception) {
+        Log.w(LOG_TAG, "Couldn't read /proc/cpuinfo", e)
+        null
+    }
+}

examples/whisper.android/app/src/main/jni/whisper/Android.mk

@@ -1,22 +1,15 @@
 LOCAL_PATH := $(call my-dir)
 include $(CLEAR_VARS)
-WHISPER_LIB_DIR := $(LOCAL_PATH)/../../../../../../../
-LOCAL_LDLIBS    := -llog
 LOCAL_MODULE    := libwhisper
+include $(LOCAL_PATH)/Whisper.mk
+include $(BUILD_SHARED_LIBRARY)
 
-# Make the final output library smaller by only keeping the symbols referenced from the app.
-ifneq ($(APP_OPTIM),debug)
-    LOCAL_CFLAGS += -fvisibility=hidden -fvisibility-inlines-hidden
-    LOCAL_CFLAGS += -ffunction-sections -fdata-sections
-    LOCAL_LDFLAGS += -Wl,--gc-sections
-    LOCAL_LDFLAGS += -Wl,--exclude-libs,ALL
-    LOCAL_LDFLAGS += -flto
-endif
-
-LOCAL_CFLAGS    += -DSTDC_HEADERS -std=c11 -I $(WHISPER_LIB_DIR)
-LOCAL_CPPFLAGS  += -std=c++11
-LOCAL_SRC_FILES := $(WHISPER_LIB_DIR)/ggml.c \
-                   $(WHISPER_LIB_DIR)/whisper.cpp \
-                   $(LOCAL_PATH)/jni.c
-
-include $(BUILD_SHARED_LIBRARY)
+ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)
+	include $(CLEAR_VARS)
+	LOCAL_MODULE    := libwhisper_vfpv4
+	include $(LOCAL_PATH)/Whisper.mk
+	# Allow building NEON FMA code.
+	# https://android.googlesource.com/platform/ndk/+/master/sources/android/cpufeatures/cpu-features.h
+	LOCAL_CFLAGS += -mfpu=neon-vfpv4
+	include $(BUILD_SHARED_LIBRARY)
+endif

examples/whisper.android/app/src/main/jni/whisper/Whisper.mk

@@ -0,0 +1,18 @@
+WHISPER_LIB_DIR := $(LOCAL_PATH)/../../../../../../../
+LOCAL_LDLIBS    := -landroid -llog
+
+# Make the final output library smaller by only keeping the symbols referenced from the app.
+ifneq ($(APP_OPTIM),debug)
+    LOCAL_CFLAGS += -O3
+    LOCAL_CFLAGS += -fvisibility=hidden -fvisibility-inlines-hidden
+    LOCAL_CFLAGS += -ffunction-sections -fdata-sections
+    LOCAL_LDFLAGS += -Wl,--gc-sections
+    LOCAL_LDFLAGS += -Wl,--exclude-libs,ALL
+    LOCAL_LDFLAGS += -flto
+endif
+
+LOCAL_CFLAGS    += -DSTDC_HEADERS -std=c11 -I $(WHISPER_LIB_DIR)
+LOCAL_CPPFLAGS  += -std=c++11
+LOCAL_SRC_FILES := $(WHISPER_LIB_DIR)/ggml.c \
+                   $(WHISPER_LIB_DIR)/whisper.cpp \
+                   $(LOCAL_PATH)/jni.c

examples/whisper.android/app/src/main/jni/whisper/jni.c

@@ -1,13 +1,17 @@
 #include <jni.h>
+#include <android/asset_manager.h>
+#include <android/asset_manager_jni.h>
 #include <android/log.h>
 #include <stdlib.h>
 #include <sys/sysinfo.h>
+#include <string.h>
 #include "whisper.h"
 
 #define UNUSED(x) (void)(x)
 #define TAG "JNI"
 
 #define LOGI(...) __android_log_print(ANDROID_LOG_INFO,     TAG, __VA_ARGS__)
+#define LOGW(...) __android_log_print(ANDROID_LOG_WARN,     TAG, __VA_ARGS__)
 
 static inline int min(int a, int b) {
     return (a < b) ? a : b;
@@ -17,13 +21,132 @@ static inline int max(int a, int b) {
     return (a > b) ? a : b;
 }
 
+struct input_stream_context {
+    size_t offset;
+    JNIEnv * env;
+    jobject thiz;
+    jobject input_stream;
+
+    jmethodID mid_available;
+    jmethodID mid_read;
+};
+
+size_t inputStreamRead(void * ctx, void * output, size_t read_size) {
+    struct input_stream_context* is = (struct input_stream_context*)ctx;
+
+    jint avail_size = (*is->env)->CallIntMethod(is->env, is->input_stream, is->mid_available);
+    jint size_to_copy = read_size < avail_size ? (jint)read_size : avail_size;
+
+    jbyteArray byte_array = (*is->env)->NewByteArray(is->env, size_to_copy);
+
+    jint n_read = (*is->env)->CallIntMethod(is->env, is->input_stream, is->mid_read, byte_array, 0, size_to_copy);
+
+    if (size_to_copy != read_size || size_to_copy != n_read) {
+        LOGI("Insufficient Read: Req=%zu, ToCopy=%d, Available=%d", read_size, size_to_copy, n_read);
+    }
+
+    jbyte* byte_array_elements = (*is->env)->GetByteArrayElements(is->env, byte_array, NULL);
+    memcpy(output, byte_array_elements, size_to_copy);
+    (*is->env)->ReleaseByteArrayElements(is->env, byte_array, byte_array_elements, JNI_ABORT);
+
+    (*is->env)->DeleteLocalRef(is->env, byte_array);
+
+    is->offset += size_to_copy;
+
+    return size_to_copy;
+}
+bool inputStreamEof(void * ctx) {
+    struct input_stream_context* is = (struct input_stream_context*)ctx;
+
+    jint result = (*is->env)->CallIntMethod(is->env, is->input_stream, is->mid_available);
+    return result <= 0;
+}
+void inputStreamClose(void * ctx) {
+
+}
+
+JNIEXPORT jlong JNICALL
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_initContextFromInputStream(
+        JNIEnv *env, jobject thiz, jobject input_stream) {
+    UNUSED(thiz);
+
+    struct whisper_context *context = NULL;
+    struct whisper_model_loader loader = {};
+    struct input_stream_context inp_ctx = {};
+
+    inp_ctx.offset = 0;
+    inp_ctx.env = env;
+    inp_ctx.thiz = thiz;
+    inp_ctx.input_stream = input_stream;
+
+    jclass cls = (*env)->GetObjectClass(env, input_stream);
+    inp_ctx.mid_available = (*env)->GetMethodID(env, cls, "available", "()I");
+    inp_ctx.mid_read = (*env)->GetMethodID(env, cls, "read", "([BII)I");
+
+    loader.context = &inp_ctx;
+    loader.read = inputStreamRead;
+    loader.eof = inputStreamEof;
+    loader.close = inputStreamClose;
+
+    loader.eof(loader.context);
+
+    context = whisper_init(&loader);
+    return (jlong) context;
+}
+
+static size_t asset_read(void *ctx, void *output, size_t read_size) {
+    return AAsset_read((AAsset *) ctx, output, read_size);
+}
+
+static bool asset_is_eof(void *ctx) {
+    return AAsset_getRemainingLength64((AAsset *) ctx) <= 0;
+}
+
+static void asset_close(void *ctx) {
+    AAsset_close((AAsset *) ctx);
+}
+
+static struct whisper_context *whisper_init_from_asset(
+        JNIEnv *env,
+        jobject assetManager,
+        const char *asset_path
+) {
+    LOGI("Loading model from asset '%s'\n", asset_path);
+    AAssetManager *asset_manager = AAssetManager_fromJava(env, assetManager);
+    AAsset *asset = AAssetManager_open(asset_manager, asset_path, AASSET_MODE_STREAMING);
+    if (!asset) {
+        LOGW("Failed to open '%s'\n", asset_path);
+        return NULL;
+    }
+
+    whisper_model_loader loader = {
+            .context = asset,
+            .read = &asset_read,
+            .eof = &asset_is_eof,
+            .close = &asset_close
+    };
+
+    return whisper_init(&loader);
+}
+
+JNIEXPORT jlong JNICALL
+Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_initContextFromAsset(
+        JNIEnv *env, jobject thiz, jobject assetManager, jstring asset_path_str) {
+    UNUSED(thiz);
+    struct whisper_context *context = NULL;
+    const char *asset_path_chars = (*env)->GetStringUTFChars(env, asset_path_str, NULL);
+    context = whisper_init_from_asset(env, assetManager, asset_path_chars);
+    (*env)->ReleaseStringUTFChars(env, asset_path_str, asset_path_chars);
+    return (jlong) context;
+}
+
 JNIEXPORT jlong JNICALL
 Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_initContext(
         JNIEnv *env, jobject thiz, jstring model_path_str) {
     UNUSED(thiz);
     struct whisper_context *context = NULL;
     const char *model_path_chars = (*env)->GetStringUTFChars(env, model_path_str, NULL);
-    context = whisper_init(model_path_chars);
+    context = whisper_init_from_file(model_path_chars);
     (*env)->ReleaseStringUTFChars(env, model_path_str, model_path_chars);
     return (jlong) context;
 }

examples/whisper.android/local.properties

@@ -1,10 +0,0 @@
-## This file is automatically generated by Android Studio.
-# Do not modify this file -- YOUR CHANGES WILL BE ERASED!
-#
-# This file should *NOT* be checked into Version Control Systems,
-# as it contains information specific to your local configuration.
-#
-# Location of the SDK. This is only used by Gradle.
-# For customization when using a Version Control System, please read the
-# header note.
-sdk.dir=/Users/kevin/Library/Android/sdk

examples/whisper.objc/README.md

@@ -19,3 +19,8 @@ open whisper.cpp/examples/whisper.objc/whisper.objc.xcodeproj/
 Make sure to build the project in `Release`:
 
 <img width="947" alt="image" src="https://user-images.githubusercontent.com/1991296/197382607-9e1e6d1b-79fa-496f-9d16-b71dc1535701.png">
+
+Also, don't forget to add the `-DGGML_USE_ACCELERATE` compiler flag in Build Phases.
+This can significantly improve the performance of the transcription:
+
+<img width="1072" alt="image" src="https://user-images.githubusercontent.com/1991296/208511239-8d7cdbd1-aa48-41b5-becd-ca288d53cc07.png">

examples/whisper.objc/whisper.objc/ViewController.m

@@ -61,7 +61,7 @@ void AudioInputCallback(void * inUserData,
         NSLog(@"Loading model from %@", modelPath);
 
         // create ggml context
-        stateInp.ctx = whisper_init([modelPath UTF8String]);
+        stateInp.ctx = whisper_init_from_file([modelPath UTF8String]);
 
         // check if the model was loaded successfully
         if (stateInp.ctx == NULL) {

examples/whisper.swiftui/README.md

@@ -0,0 +1,14 @@
+A sample SwiftUI app using [whisper.cpp](https://github.com/ggerganov/whisper.cpp/) to do voice-to-text transcriptions.
+See also: [whisper.objc](https://github.com/ggerganov/whisper.cpp/tree/master/examples/whisper.objc).
+
+To use:
+
+1. Select a model from the [whisper.cpp repository](https://github.com/ggerganov/whisper.cpp/tree/master/models).[^1]
+2. Add the model to "whisper.swiftui.demo/Resources/models" via Xcode.
+3. Select a sample audio file (for example, [jfk.wav](https://github.com/ggerganov/whisper.cpp/raw/master/samples/jfk.wav)).
+4. Add the model to "whisper.swiftui.demo/Resources/samples" via Xcode.
+5. Select the "release" build configuration under "Run", then deploy and run to your device.
+
+[^1]: I recommend the tiny, base or small models for running on an iOS device.
+
+![image](https://user-images.githubusercontent.com/1991296/212539216-0aef65e4-f882-480a-8358-0f816838fd52.png)

examples/whisper.swiftui/whisper.cpp.swift/LibWhisper.swift

@@ -0,0 +1,70 @@
+import Foundation
+
+enum WhisperError: Error {
+    case couldNotInitializeContext
+}
+
+// Meet Whisper C++ constraint: Don't access from more than one thread at a time.
+actor WhisperContext {
+    private var context: OpaquePointer
+    
+    init(context: OpaquePointer) {
+        self.context = context
+    }
+    
+    deinit {
+        whisper_free(context)
+    }
+    
+    func fullTranscribe(samples: [Float]) {
+        // Leave 2 processors free (i.e. the high-efficiency cores).
+        let maxThreads = max(1, min(8, cpuCount() - 2))
+        print("Selecting \(maxThreads) threads")
+        var params = whisper_full_default_params(WHISPER_SAMPLING_GREEDY)
+        "en".withCString { en in
+            // Adapted from whisper.objc
+            params.print_realtime = true
+            params.print_progress = false
+            params.print_timestamps = true
+            params.print_special = false
+            params.translate = false
+            params.language = en
+            params.n_threads = Int32(maxThreads)
+            params.offset_ms = 0
+            params.no_context = true
+            params.single_segment = false
+            
+            whisper_reset_timings(context)
+            print("About to run whisper_full")
+            samples.withUnsafeBufferPointer { samples in
+                if (whisper_full(context, params, samples.baseAddress, Int32(samples.count)) != 0) {
+                    print("Failed to run the model")
+                } else {
+                    whisper_print_timings(context)
+                }
+            }
+        }
+    }
+    
+    func getTranscription() -> String {
+        var transcription = ""
+        for i in 0..<whisper_full_n_segments(context) {
+            transcription += String.init(cString: whisper_full_get_segment_text(context, i))
+        }
+        return transcription
+    }
+    
+    static func createContext(path: String) throws -> WhisperContext {
+        let context = whisper_init_from_file(path)
+        if let context {
+            return WhisperContext(context: context)
+        } else {
+            print("Couldn't load model at \(path)")
+            throw WhisperError.couldNotInitializeContext
+        }
+    }
+}
+
+fileprivate func cpuCount() -> Int {
+    ProcessInfo.processInfo.processorCount
+}

examples/whisper.swiftui/whisper.cpp.swift/WhisperCppDemo-Bridging-Header.h

@@ -0,0 +1,4 @@
+//
+//  Use this file to import your target's public headers that you would like to expose to Swift.
+//
+#import "whisper.h"

examples/whisper.swiftui/whisper.swiftui.demo/Models/WhisperState.swift

@@ -0,0 +1,162 @@
+import Foundation
+import SwiftUI
+import AVFoundation
+
+@MainActor
+class WhisperState: NSObject, ObservableObject, AVAudioRecorderDelegate {
+    @Published var isModelLoaded = false
+    @Published var messageLog = ""
+    @Published var canTranscribe = false
+    @Published var isRecording = false
+    
+    private var whisperContext: WhisperContext?
+    private let recorder = Recorder()
+    private var recordedFile: URL? = nil
+    private var audioPlayer: AVAudioPlayer?
+    
+    private var modelUrl: URL? {
+        Bundle.main.url(forResource: "ggml-tiny.en", withExtension: "bin", subdirectory: "models")
+    }
+    
+    private var sampleUrl: URL? {
+        Bundle.main.url(forResource: "jfk", withExtension: "wav", subdirectory: "samples")
+    }
+    
+    private enum LoadError: Error {
+        case couldNotLocateModel
+    }
+    
+    override init() {
+        super.init()
+        do {
+            try loadModel()
+            canTranscribe = true
+        } catch {
+            print(error.localizedDescription)
+            messageLog += "\(error.localizedDescription)\n"
+        }
+    }
+    
+    private func loadModel() throws {
+        messageLog += "Loading model...\n"
+        if let modelUrl {
+            whisperContext = try WhisperContext.createContext(path: modelUrl.path())
+            messageLog += "Loaded model \(modelUrl.lastPathComponent)\n"
+        } else {
+            messageLog += "Could not locate model\n"
+        }
+    }
+    
+    func transcribeSample() async {
+        if let sampleUrl {
+            await transcribeAudio(sampleUrl)
+        } else {
+            messageLog += "Could not locate sample\n"
+        }
+    }
+    
+    private func transcribeAudio(_ url: URL) async {
+        if (!canTranscribe) {
+            return
+        }
+        guard let whisperContext else {
+            return
+        }
+        
+        do {
+            canTranscribe = false
+            messageLog += "Reading wave samples...\n"
+            let data = try readAudioSamples(url)
+            messageLog += "Transcribing data...\n"
+            await whisperContext.fullTranscribe(samples: data)
+            let text = await whisperContext.getTranscription()
+            messageLog += "Done: \(text)\n"
+        } catch {
+            print(error.localizedDescription)
+            messageLog += "\(error.localizedDescription)\n"
+        }
+        
+        canTranscribe = true
+    }
+    
+    private func readAudioSamples(_ url: URL) throws -> [Float] {
+        stopPlayback()
+        try startPlayback(url)
+        return try decodeWaveFile(url)
+    }
+    
+    func toggleRecord() async {
+        if isRecording {
+            await recorder.stopRecording()
+            isRecording = false
+            if let recordedFile {
+                await transcribeAudio(recordedFile)
+            }
+        } else {
+            requestRecordPermission { granted in
+                if granted {
+                    Task {
+                        do {
+                            self.stopPlayback()
+                            let file = try FileManager.default.url(for: .documentDirectory, in: .userDomainMask, appropriateFor: nil, create: true)
+                                .appending(path: "output.wav")
+                            try await self.recorder.startRecording(toOutputFile: file, delegate: self)
+                            self.isRecording = true
+                            self.recordedFile = file
+                        } catch {
+                            print(error.localizedDescription)
+                            self.messageLog += "\(error.localizedDescription)\n"
+                            self.isRecording = false
+                        }
+                    }
+                }
+            }
+        }
+    }
+    
+    private func requestRecordPermission(response: @escaping (Bool) -> Void) {
+#if os(macOS)
+        response(true)
+#else
+        AVAudioSession.sharedInstance().requestRecordPermission { granted in
+            response(granted)
+        }
+#endif
+    }
+    
+    private func startPlayback(_ url: URL) throws {
+        audioPlayer = try AVAudioPlayer(contentsOf: url)
+        audioPlayer?.play()
+    }
+    
+    private func stopPlayback() {
+        audioPlayer?.stop()
+        audioPlayer = nil
+    }
+    
+    // MARK: AVAudioRecorderDelegate
+    
+    nonisolated func audioRecorderEncodeErrorDidOccur(_ recorder: AVAudioRecorder, error: Error?) {
+        if let error {
+            Task {
+                await handleRecError(error)
+            }
+        }
+    }
+    
+    private func handleRecError(_ error: Error) {
+        print(error.localizedDescription)
+        messageLog += "\(error.localizedDescription)\n"
+        isRecording = false
+    }
+    
+    nonisolated func audioRecorderDidFinishRecording(_ recorder: AVAudioRecorder, successfully flag: Bool) {
+        Task {
+            await onDidFinishRecording()
+        }
+    }
+    
+    private func onDidFinishRecording() {
+        isRecording = false
+    }
+}

examples/whisper.swiftui/whisper.swiftui.demo/Supporting files/Assets.xcassets/AccentColor.colorset/Contents.json

@@ -0,0 +1,11 @@
+{
+  "colors" : [
+    {
+      "idiom" : "universal"
+    }
+  ],
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}

examples/whisper.swiftui/whisper.swiftui.demo/Supporting files/Assets.xcassets/AppIcon.appiconset/Contents.json

@@ -0,0 +1,63 @@
+{
+  "images" : [
+    {
+      "idiom" : "universal",
+      "platform" : "ios",
+      "size" : "1024x1024"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "1x",
+      "size" : "16x16"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "2x",
+      "size" : "16x16"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "1x",
+      "size" : "32x32"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "2x",
+      "size" : "32x32"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "1x",
+      "size" : "128x128"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "2x",
+      "size" : "128x128"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "1x",
+      "size" : "256x256"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "2x",
+      "size" : "256x256"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "1x",
+      "size" : "512x512"
+    },
+    {
+      "idiom" : "mac",
+      "scale" : "2x",
+      "size" : "512x512"
+    }
+  ],
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}

examples/whisper.swiftui/whisper.swiftui.demo/Supporting files/Assets.xcassets/Contents.json

@@ -0,0 +1,6 @@
+{
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}

examples/whisper.swiftui/whisper.swiftui.demo/Supporting files/Preview Content/Preview Assets.xcassets/Contents.json

@@ -0,0 +1,6 @@
+{
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}

examples/whisper.swiftui/whisper.swiftui.demo/Supporting files/WhisperCppDemo.entitlements

@@ -0,0 +1,12 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+	<key>com.apple.security.app-sandbox</key>
+	<true/>
+	<key>com.apple.security.device.audio-input</key>
+	<true/>
+	<key>com.apple.security.files.user-selected.read-only</key>
+	<true/>
+</dict>
+</plist>

examples/whisper.swiftui/whisper.swiftui.demo/UI/ContentView.swift

@@ -0,0 +1,43 @@
+import SwiftUI
+import AVFoundation
+
+struct ContentView: View {
+    @StateObject var whisperState = WhisperState()
+    
+    var body: some View {
+        NavigationStack {
+            VStack {
+                HStack {
+                    Button("Transcribe", action: {
+                        Task {
+                            await whisperState.transcribeSample()
+                        }
+                    })
+                    .buttonStyle(.bordered)
+                    .disabled(!whisperState.canTranscribe)
+                    
+                    Button(whisperState.isRecording ? "Stop recording" : "Start recording", action: {
+                        Task {
+                            await whisperState.toggleRecord()
+                        }
+                    })
+                    .buttonStyle(.bordered)
+                    .disabled(!whisperState.canTranscribe)
+                }
+                
+                ScrollView {
+                    Text(verbatim: whisperState.messageLog)
+                        .frame(maxWidth: .infinity, alignment: .leading)
+                }
+            }
+            .navigationTitle("Whisper SwiftUI Demo")
+            .padding()
+        }
+    }
+}
+
+struct ContentView_Previews: PreviewProvider {
+    static var previews: some View {
+        ContentView()
+    }
+}

examples/whisper.swiftui/whisper.swiftui.demo/Utils/Recorder.swift

@@ -0,0 +1,35 @@
+import Foundation
+import AVFoundation
+
+actor Recorder {
+    private var recorder: AVAudioRecorder?
+    
+    enum RecorderError: Error {
+        case couldNotStartRecording
+    }
+    
+    func startRecording(toOutputFile url: URL, delegate: AVAudioRecorderDelegate?) throws {
+        let recordSettings: [String : Any] = [
+            AVFormatIDKey: Int(kAudioFormatLinearPCM),
+            AVSampleRateKey: 16000.0,
+            AVNumberOfChannelsKey: 1,
+            AVEncoderAudioQualityKey: AVAudioQuality.high.rawValue
+        ]
+#if !os(macOS)
+        let session = AVAudioSession.sharedInstance()
+        try session.setCategory(.playAndRecord, mode: .default)
+#endif
+        let recorder = try AVAudioRecorder(url: url, settings: recordSettings)
+        recorder.delegate = delegate
+        if recorder.record() == false {
+            print("Could not start recording")
+            throw RecorderError.couldNotStartRecording
+        }
+        self.recorder = recorder
+    }
+    
+    func stopRecording() {
+        recorder?.stop()
+        recorder = nil
+    }
+}

examples/whisper.swiftui/whisper.swiftui.demo/Utils/RiffWaveUtils.swift

@@ -0,0 +1,12 @@
+import Foundation
+
+func decodeWaveFile(_ url: URL) throws -> [Float] {
+    let data = try Data(contentsOf: url)
+    let floats = stride(from: 44, to: data.count, by: 2).map {
+        return data[$0..<$0 + 2].withUnsafeBytes {
+            let short = Int16(littleEndian: $0.load(as: Int16.self))
+            return max(-1.0, min(Float(short) / 32767.0, 1.0))
+        }
+    }
+    return floats
+}

examples/whisper.swiftui/whisper.swiftui.demo/WhisperCppDemoApp.swift

@@ -0,0 +1,10 @@
+import SwiftUI
+
+@main
+struct WhisperCppDemoApp: App {
+    var body: some Scene {
+        WindowGroup {
+            ContentView()
+        }
+    }
+}

examples/whisper.swiftui/whisper.swiftui.xcodeproj/.gitignore

@@ -0,0 +1 @@
+xcuserdata/

examples/whisper.swiftui/whisper.swiftui.xcodeproj/project.pbxproj

@@ -0,0 +1,468 @@
+// !$*UTF8*$!
+{
+	archiveVersion = 1;
+	classes = {
+	};
+	objectVersion = 56;
+	objects = {
+
+/* Begin PBXBuildFile section */
+		0A8E49002954B3F100704C1B /* README.md in Resources */ = {isa = PBXBuildFile; fileRef = 0A8E48FF2954B3F100704C1B /* README.md */; };
+		0AA751482953AC2E001EE061 /* samples in Resources */ = {isa = PBXBuildFile; fileRef = 0AA751462953AC2E001EE061 /* samples */; };
+		0AA751492953AC2E001EE061 /* models in Resources */ = {isa = PBXBuildFile; fileRef = 0AA751472953AC2E001EE061 /* models */; };
+		0AA7514C2953B569001EE061 /* RiffWaveUtils.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AA7514B2953B569001EE061 /* RiffWaveUtils.swift */; };
+		0AA7514E2953D958001EE061 /* Recorder.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AA7514D2953D958001EE061 /* Recorder.swift */; };
+		0AAC5D9B29539CCF003032C3 /* WhisperCppDemoApp.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5D9A29539CCF003032C3 /* WhisperCppDemoApp.swift */; };
+		0AAC5D9D29539CCF003032C3 /* ContentView.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5D9C29539CCF003032C3 /* ContentView.swift */; };
+		0AAC5D9F29539CD0003032C3 /* Assets.xcassets in Resources */ = {isa = PBXBuildFile; fileRef = 0AAC5D9E29539CD0003032C3 /* Assets.xcassets */; };
+		0AAC5DA329539CD0003032C3 /* Preview Assets.xcassets in Resources */ = {isa = PBXBuildFile; fileRef = 0AAC5DA229539CD0003032C3 /* Preview Assets.xcassets */; };
+		0AAC5DCB29539EB1003032C3 /* whisper.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DC729539EB0003032C3 /* whisper.cpp */; settings = {COMPILER_FLAGS = "-Wno-shorten-64-to-32"; }; };
+		0AAC5DCC29539EB1003032C3 /* ggml.c in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DC929539EB0003032C3 /* ggml.c */; settings = {COMPILER_FLAGS = "-DGGML_USE_ACCELERATE -Wno-shorten-64-to-32"; }; };
+		0AAC5DCE2953A05C003032C3 /* WhisperState.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DCD2953A05C003032C3 /* WhisperState.swift */; };
+		0AAC5DD12953A394003032C3 /* LibWhisper.swift in Sources */ = {isa = PBXBuildFile; fileRef = 0AAC5DD02953A394003032C3 /* LibWhisper.swift */; };
+/* End PBXBuildFile section */
+
+/* Begin PBXFileReference section */
+		0A8E48FF2954B3F100704C1B /* README.md */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = net.daringfireball.markdown; path = README.md; sourceTree = "<group>"; };
+		0AA751462953AC2E001EE061 /* samples */ = {isa = PBXFileReference; lastKnownFileType = folder; path = samples; sourceTree = "<group>"; };
+		0AA751472953AC2E001EE061 /* models */ = {isa = PBXFileReference; lastKnownFileType = folder; path = models; sourceTree = "<group>"; };
+		0AA7514B2953B569001EE061 /* RiffWaveUtils.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = RiffWaveUtils.swift; sourceTree = "<group>"; };
+		0AA7514D2953D958001EE061 /* Recorder.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = Recorder.swift; sourceTree = "<group>"; };
+		0AAC5D9729539CCF003032C3 /* whisper.swiftui.app */ = {isa = PBXFileReference; explicitFileType = wrapper.application; includeInIndex = 0; path = whisper.swiftui.app; sourceTree = BUILT_PRODUCTS_DIR; };
+		0AAC5D9A29539CCF003032C3 /* WhisperCppDemoApp.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = WhisperCppDemoApp.swift; sourceTree = "<group>"; };
+		0AAC5D9C29539CCF003032C3 /* ContentView.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ContentView.swift; sourceTree = "<group>"; };
+		0AAC5D9E29539CD0003032C3 /* Assets.xcassets */ = {isa = PBXFileReference; lastKnownFileType = folder.assetcatalog; path = Assets.xcassets; sourceTree = "<group>"; };
+		0AAC5DA029539CD0003032C3 /* WhisperCppDemo.entitlements */ = {isa = PBXFileReference; lastKnownFileType = text.plist.entitlements; path = WhisperCppDemo.entitlements; sourceTree = "<group>"; };
+		0AAC5DA229539CD0003032C3 /* Preview Assets.xcassets */ = {isa = PBXFileReference; lastKnownFileType = folder.assetcatalog; path = "Preview Assets.xcassets"; sourceTree = "<group>"; };
+		0AAC5DC629539EAF003032C3 /* WhisperCppDemo-Bridging-Header.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = "WhisperCppDemo-Bridging-Header.h"; sourceTree = "<group>"; };
+		0AAC5DC729539EB0003032C3 /* whisper.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = whisper.cpp; sourceTree = "<group>"; };
+		0AAC5DC829539EB0003032C3 /* whisper.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = whisper.h; sourceTree = "<group>"; };
+		0AAC5DC929539EB0003032C3 /* ggml.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ggml.c; sourceTree = "<group>"; };
+		0AAC5DCA29539EB0003032C3 /* ggml.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ggml.h; sourceTree = "<group>"; };
+		0AAC5DCD2953A05C003032C3 /* WhisperState.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = WhisperState.swift; sourceTree = "<group>"; };
+		0AAC5DD02953A394003032C3 /* LibWhisper.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = LibWhisper.swift; sourceTree = "<group>"; };
+/* End PBXFileReference section */
+
+/* Begin PBXFrameworksBuildPhase section */
+		0AAC5D9429539CCF003032C3 /* Frameworks */ = {
+			isa = PBXFrameworksBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXFrameworksBuildPhase section */
+
+/* Begin PBXGroup section */
+		0AA7513F2953AB32001EE061 /* Models */ = {
+			isa = PBXGroup;
+			children = (
+				0AAC5DCD2953A05C003032C3 /* WhisperState.swift */,
+			);
+			path = Models;
+			sourceTree = "<group>";
+		};
+		0AA751402953ABA6001EE061 /* Resources */ = {
+			isa = PBXGroup;
+			children = (
+				0AA751472953AC2E001EE061 /* models */,
+				0AA751462953AC2E001EE061 /* samples */,
+			);
+			path = Resources;
+			sourceTree = "<group>";
+		};
+		0AA7514A2953B561001EE061 /* Utils */ = {
+			isa = PBXGroup;
+			children = (
+				0AA7514B2953B569001EE061 /* RiffWaveUtils.swift */,
+				0AA7514D2953D958001EE061 /* Recorder.swift */,
+			);
+			path = Utils;
+			sourceTree = "<group>";
+		};
+		0AAC5D8E29539CCF003032C3 = {
+			isa = PBXGroup;
+			children = (
+				0A8E48FF2954B3F100704C1B /* README.md */,
+				0AAC5DC529539E89003032C3 /* whisper.cpp */,
+				0AAC5DCF2953A36C003032C3 /* whisper.cpp.swift */,
+				0AAC5D9929539CCF003032C3 /* whisper.swiftui.demo */,
+				0AAC5D9829539CCF003032C3 /* Products */,
+			);
+			sourceTree = "<group>";
+		};
+		0AAC5D9829539CCF003032C3 /* Products */ = {
+			isa = PBXGroup;
+			children = (
+				0AAC5D9729539CCF003032C3 /* whisper.swiftui.app */,
+			);
+			name = Products;
+			sourceTree = "<group>";
+		};
+		0AAC5D9929539CCF003032C3 /* whisper.swiftui.demo */ = {
+			isa = PBXGroup;
+			children = (
+				0AA7514A2953B561001EE061 /* Utils */,
+				0AA751402953ABA6001EE061 /* Resources */,
+				0AA7513F2953AB32001EE061 /* Models */,
+				0AAC5DD32953A9ED003032C3 /* Supporting files */,
+				0AAC5DD22953A9E3003032C3 /* UI */,
+				0AAC5D9A29539CCF003032C3 /* WhisperCppDemoApp.swift */,
+			);
+			path = whisper.swiftui.demo;
+			sourceTree = "<group>";
+		};
+		0AAC5DA129539CD0003032C3 /* Preview Content */ = {
+			isa = PBXGroup;
+			children = (
+				0AAC5DA229539CD0003032C3 /* Preview Assets.xcassets */,
+			);
+			name = "Preview Content";
+			path = "../Preview Content";
+			sourceTree = "<group>";
+		};
+		0AAC5DC529539E89003032C3 /* whisper.cpp */ = {
+			isa = PBXGroup;
+			children = (
+				0AAC5DC929539EB0003032C3 /* ggml.c */,
+				0AAC5DCA29539EB0003032C3 /* ggml.h */,
+				0AAC5DC729539EB0003032C3 /* whisper.cpp */,
+				0AAC5DC829539EB0003032C3 /* whisper.h */,
+			);
+			name = whisper.cpp;
+			path = ../..;
+			sourceTree = "<group>";
+		};
+		0AAC5DCF2953A36C003032C3 /* whisper.cpp.swift */ = {
+			isa = PBXGroup;
+			children = (
+				0AAC5DC629539EAF003032C3 /* WhisperCppDemo-Bridging-Header.h */,
+				0AAC5DD02953A394003032C3 /* LibWhisper.swift */,
+			);
+			path = whisper.cpp.swift;
+			sourceTree = "<group>";
+		};
+		0AAC5DD22953A9E3003032C3 /* UI */ = {
+			isa = PBXGroup;
+			children = (
+				0AAC5D9C29539CCF003032C3 /* ContentView.swift */,
+			);
+			path = UI;
+			sourceTree = "<group>";
+		};
+		0AAC5DD32953A9ED003032C3 /* Supporting files */ = {
+			isa = PBXGroup;
+			children = (
+				0AAC5D9E29539CD0003032C3 /* Assets.xcassets */,
+				0AAC5DA029539CD0003032C3 /* WhisperCppDemo.entitlements */,
+				0AAC5DA129539CD0003032C3 /* Preview Content */,
+			);
+			path = "Supporting files";
+			sourceTree = "<group>";
+		};
+/* End PBXGroup section */
+
+/* Begin PBXNativeTarget section */
+		0AAC5D9629539CCF003032C3 /* whisper.swiftui */ = {
+			isa = PBXNativeTarget;
+			buildConfigurationList = 0AAC5DBC29539CD0003032C3 /* Build configuration list for PBXNativeTarget "whisper.swiftui" */;
+			buildPhases = (
+				0AAC5D9329539CCF003032C3 /* Sources */,
+				0AAC5D9429539CCF003032C3 /* Frameworks */,
+				0AAC5D9529539CCF003032C3 /* Resources */,
+			);
+			buildRules = (
+			);
+			dependencies = (
+			);
+			name = whisper.swiftui;
+			productName = WhisperCppDemo;
+			productReference = 0AAC5D9729539CCF003032C3 /* whisper.swiftui.app */;
+			productType = "com.apple.product-type.application";
+		};
+/* End PBXNativeTarget section */
+
+/* Begin PBXProject section */
+		0AAC5D8F29539CCF003032C3 /* Project object */ = {
+			isa = PBXProject;
+			attributes = {
+				BuildIndependentTargetsInParallel = 1;
+				LastSwiftUpdateCheck = 1410;
+				LastUpgradeCheck = 1410;
+				TargetAttributes = {
+					0AAC5D9629539CCF003032C3 = {
+						CreatedOnToolsVersion = 14.1;
+						LastSwiftMigration = 1410;
+					};
+				};
+			};
+			buildConfigurationList = 0AAC5D9229539CCF003032C3 /* Build configuration list for PBXProject "whisper.swiftui" */;
+			compatibilityVersion = "Xcode 14.0";
+			developmentRegion = en;
+			hasScannedForEncodings = 0;
+			knownRegions = (
+				en,
+				Base,
+			);
+			mainGroup = 0AAC5D8E29539CCF003032C3;
+			productRefGroup = 0AAC5D9829539CCF003032C3 /* Products */;
+			projectDirPath = "";
+			projectRoot = "";
+			targets = (
+				0AAC5D9629539CCF003032C3 /* whisper.swiftui */,
+			);
+		};
+/* End PBXProject section */
+
+/* Begin PBXResourcesBuildPhase section */
+		0AAC5D9529539CCF003032C3 /* Resources */ = {
+			isa = PBXResourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				0AA751482953AC2E001EE061 /* samples in Resources */,
+				0AAC5DA329539CD0003032C3 /* Preview Assets.xcassets in Resources */,
+				0A8E49002954B3F100704C1B /* README.md in Resources */,
+				0AA751492953AC2E001EE061 /* models in Resources */,
+				0AAC5D9F29539CD0003032C3 /* Assets.xcassets in Resources */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXResourcesBuildPhase section */
+
+/* Begin PBXSourcesBuildPhase section */
+		0AAC5D9329539CCF003032C3 /* Sources */ = {
+			isa = PBXSourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				0AAC5D9D29539CCF003032C3 /* ContentView.swift in Sources */,
+				0AAC5D9B29539CCF003032C3 /* WhisperCppDemoApp.swift in Sources */,
+				0AAC5DCC29539EB1003032C3 /* ggml.c in Sources */,
+				0AAC5DCE2953A05C003032C3 /* WhisperState.swift in Sources */,
+				0AAC5DD12953A394003032C3 /* LibWhisper.swift in Sources */,
+				0AA7514C2953B569001EE061 /* RiffWaveUtils.swift in Sources */,
+				0AAC5DCB29539EB1003032C3 /* whisper.cpp in Sources */,
+				0AA7514E2953D958001EE061 /* Recorder.swift in Sources */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXSourcesBuildPhase section */
+
+/* Begin XCBuildConfiguration section */
+		0AAC5DBA29539CD0003032C3 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ALWAYS_SEARCH_USER_PATHS = NO;
+				CLANG_ANALYZER_NONNULL = YES;
+				CLANG_ANALYZER_NUMBER_OBJECT_CONVERSION = YES_AGGRESSIVE;
+				CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
+				CLANG_ENABLE_MODULES = YES;
+				CLANG_ENABLE_OBJC_ARC = YES;
+				CLANG_ENABLE_OBJC_WEAK = YES;
+				CLANG_WARN_BLOCK_CAPTURE_AUTORELEASING = YES;
+				CLANG_WARN_BOOL_CONVERSION = YES;
+				CLANG_WARN_COMMA = YES;
+				CLANG_WARN_CONSTANT_CONVERSION = YES;
+				CLANG_WARN_DEPRECATED_OBJC_IMPLEMENTATIONS = YES;
+				CLANG_WARN_DIRECT_OBJC_ISA_USAGE = YES_ERROR;
+				CLANG_WARN_DOCUMENTATION_COMMENTS = YES;
+				CLANG_WARN_EMPTY_BODY = YES;
+				CLANG_WARN_ENUM_CONVERSION = YES;
+				CLANG_WARN_INFINITE_RECURSION = YES;
+				CLANG_WARN_INT_CONVERSION = YES;
+				CLANG_WARN_NON_LITERAL_NULL_CONVERSION = YES;
+				CLANG_WARN_OBJC_IMPLICIT_RETAIN_SELF = YES;
+				CLANG_WARN_OBJC_LITERAL_CONVERSION = YES;
+				CLANG_WARN_OBJC_ROOT_CLASS = YES_ERROR;
+				CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
+				CLANG_WARN_RANGE_LOOP_ANALYSIS = YES;
+				CLANG_WARN_STRICT_PROTOTYPES = YES;
+				CLANG_WARN_SUSPICIOUS_MOVE = YES;
+				CLANG_WARN_UNGUARDED_AVAILABILITY = YES_AGGRESSIVE;
+				CLANG_WARN_UNREACHABLE_CODE = YES;
+				CLANG_WARN__DUPLICATE_METHOD_MATCH = YES;
+				COPY_PHASE_STRIP = NO;
+				DEBUG_INFORMATION_FORMAT = dwarf;
+				ENABLE_STRICT_OBJC_MSGSEND = YES;
+				ENABLE_TESTABILITY = YES;
+				GCC_C_LANGUAGE_STANDARD = gnu11;
+				GCC_DYNAMIC_NO_PIC = NO;
+				GCC_NO_COMMON_BLOCKS = YES;
+				GCC_OPTIMIZATION_LEVEL = 0;
+				GCC_PREPROCESSOR_DEFINITIONS = (
+					"DEBUG=1",
+					"$(inherited)",
+				);
+				GCC_WARN_64_TO_32_BIT_CONVERSION = YES;
+				GCC_WARN_ABOUT_RETURN_TYPE = YES_ERROR;
+				GCC_WARN_UNDECLARED_SELECTOR = YES;
+				GCC_WARN_UNINITIALIZED_AUTOS = YES_AGGRESSIVE;
+				GCC_WARN_UNUSED_FUNCTION = YES;
+				GCC_WARN_UNUSED_VARIABLE = YES;
+				INFOPLIST_KEY_NSMicrophoneUsageDescription = "Needed to transcribe audio";
+				MTL_ENABLE_DEBUG_INFO = INCLUDE_SOURCE;
+				MTL_FAST_MATH = YES;
+				ONLY_ACTIVE_ARCH = YES;
+				SWIFT_ACTIVE_COMPILATION_CONDITIONS = DEBUG;
+				SWIFT_OPTIMIZATION_LEVEL = "-Onone";
+			};
+			name = Debug;
+		};
+		0AAC5DBB29539CD0003032C3 /* Release */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ALWAYS_SEARCH_USER_PATHS = NO;
+				CLANG_ANALYZER_NONNULL = YES;
+				CLANG_ANALYZER_NUMBER_OBJECT_CONVERSION = YES_AGGRESSIVE;
+				CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
+				CLANG_ENABLE_MODULES = YES;
+				CLANG_ENABLE_OBJC_ARC = YES;
+				CLANG_ENABLE_OBJC_WEAK = YES;
+				CLANG_WARN_BLOCK_CAPTURE_AUTORELEASING = YES;
+				CLANG_WARN_BOOL_CONVERSION = YES;
+				CLANG_WARN_COMMA = YES;
+				CLANG_WARN_CONSTANT_CONVERSION = YES;
+				CLANG_WARN_DEPRECATED_OBJC_IMPLEMENTATIONS = YES;
+				CLANG_WARN_DIRECT_OBJC_ISA_USAGE = YES_ERROR;
+				CLANG_WARN_DOCUMENTATION_COMMENTS = YES;
+				CLANG_WARN_EMPTY_BODY = YES;
+				CLANG_WARN_ENUM_CONVERSION = YES;
+				CLANG_WARN_INFINITE_RECURSION = YES;
+				CLANG_WARN_INT_CONVERSION = YES;
+				CLANG_WARN_NON_LITERAL_NULL_CONVERSION = YES;
+				CLANG_WARN_OBJC_IMPLICIT_RETAIN_SELF = YES;
+				CLANG_WARN_OBJC_LITERAL_CONVERSION = YES;
+				CLANG_WARN_OBJC_ROOT_CLASS = YES_ERROR;
+				CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
+				CLANG_WARN_RANGE_LOOP_ANALYSIS = YES;
+				CLANG_WARN_STRICT_PROTOTYPES = YES;
+				CLANG_WARN_SUSPICIOUS_MOVE = YES;
+				CLANG_WARN_UNGUARDED_AVAILABILITY = YES_AGGRESSIVE;
+				CLANG_WARN_UNREACHABLE_CODE = YES;
+				CLANG_WARN__DUPLICATE_METHOD_MATCH = YES;
+				COPY_PHASE_STRIP = NO;
+				DEBUG_INFORMATION_FORMAT = "dwarf-with-dsym";
+				ENABLE_NS_ASSERTIONS = NO;
+				ENABLE_STRICT_OBJC_MSGSEND = YES;
+				GCC_C_LANGUAGE_STANDARD = gnu11;
+				GCC_NO_COMMON_BLOCKS = YES;
+				GCC_WARN_64_TO_32_BIT_CONVERSION = YES;
+				GCC_WARN_ABOUT_RETURN_TYPE = YES_ERROR;
+				GCC_WARN_UNDECLARED_SELECTOR = YES;
+				GCC_WARN_UNINITIALIZED_AUTOS = YES_AGGRESSIVE;
+				GCC_WARN_UNUSED_FUNCTION = YES;
+				GCC_WARN_UNUSED_VARIABLE = YES;
+				INFOPLIST_KEY_NSMicrophoneUsageDescription = "Needed to transcribe audio";
+				MTL_ENABLE_DEBUG_INFO = NO;
+				MTL_FAST_MATH = YES;
+				SWIFT_COMPILATION_MODE = wholemodule;
+				SWIFT_OPTIMIZATION_LEVEL = "-O";
+			};
+			name = Release;
+		};
+		0AAC5DBD29539CD0003032C3 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ASSETCATALOG_COMPILER_APPICON_NAME = AppIcon;
+				ASSETCATALOG_COMPILER_GLOBAL_ACCENT_COLOR_NAME = AccentColor;
+				CLANG_ENABLE_MODULES = YES;
+				CODE_SIGN_ENTITLEMENTS = "whisper.swiftui.demo/Supporting files/WhisperCppDemo.entitlements";
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				DEVELOPMENT_ASSET_PATHS = "\"whisper.swiftui.demo/Supporting files/Preview Content\"";
+				DEVELOPMENT_TEAM = 3TZ9BM962G;
+				ENABLE_HARDENED_RUNTIME = YES;
+				ENABLE_PREVIEWS = YES;
+				GENERATE_INFOPLIST_FILE = YES;
+				"INFOPLIST_KEY_UIApplicationSceneManifest_Generation[sdk=iphoneos*]" = YES;
+				"INFOPLIST_KEY_UIApplicationSceneManifest_Generation[sdk=iphonesimulator*]" = YES;
+				"INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents[sdk=iphoneos*]" = YES;
+				"INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents[sdk=iphonesimulator*]" = YES;
+				"INFOPLIST_KEY_UILaunchScreen_Generation[sdk=iphoneos*]" = YES;
+				"INFOPLIST_KEY_UILaunchScreen_Generation[sdk=iphonesimulator*]" = YES;
+				"INFOPLIST_KEY_UIStatusBarStyle[sdk=iphoneos*]" = UIStatusBarStyleDefault;
+				"INFOPLIST_KEY_UIStatusBarStyle[sdk=iphonesimulator*]" = UIStatusBarStyleDefault;
+				INFOPLIST_KEY_UISupportedInterfaceOrientations_iPad = "UIInterfaceOrientationPortrait UIInterfaceOrientationPortraitUpsideDown UIInterfaceOrientationLandscapeLeft UIInterfaceOrientationLandscapeRight";
+				INFOPLIST_KEY_UISupportedInterfaceOrientations_iPhone = "UIInterfaceOrientationPortrait UIInterfaceOrientationLandscapeLeft UIInterfaceOrientationLandscapeRight";
+				IPHONEOS_DEPLOYMENT_TARGET = 16.1;
+				LD_RUNPATH_SEARCH_PATHS = "@executable_path/Frameworks";
+				"LD_RUNPATH_SEARCH_PATHS[sdk=macosx*]" = "@executable_path/../Frameworks";
+				MACOSX_DEPLOYMENT_TARGET = 13.0;
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = com.whispercppdemo.WhisperCppDemo;
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				SDKROOT = auto;
+				SUPPORTED_PLATFORMS = "iphoneos iphonesimulator macosx";
+				SWIFT_EMIT_LOC_STRINGS = YES;
+				SWIFT_OBJC_BRIDGING_HEADER = "whisper.cpp.swift/WhisperCppDemo-Bridging-Header.h";
+				SWIFT_OPTIMIZATION_LEVEL = "-Onone";
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+			};
+			name = Debug;
+		};
+		0AAC5DBE29539CD0003032C3 /* Release */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ASSETCATALOG_COMPILER_APPICON_NAME = AppIcon;
+				ASSETCATALOG_COMPILER_GLOBAL_ACCENT_COLOR_NAME = AccentColor;
+				CLANG_ENABLE_MODULES = YES;
+				CODE_SIGN_ENTITLEMENTS = "whisper.swiftui.demo/Supporting files/WhisperCppDemo.entitlements";
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				DEVELOPMENT_ASSET_PATHS = "\"whisper.swiftui.demo/Supporting files/Preview Content\"";
+				DEVELOPMENT_TEAM = 3TZ9BM962G;
+				ENABLE_HARDENED_RUNTIME = YES;
+				ENABLE_PREVIEWS = YES;
+				GENERATE_INFOPLIST_FILE = YES;
+				"INFOPLIST_KEY_UIApplicationSceneManifest_Generation[sdk=iphoneos*]" = YES;
+				"INFOPLIST_KEY_UIApplicationSceneManifest_Generation[sdk=iphonesimulator*]" = YES;
+				"INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents[sdk=iphoneos*]" = YES;
+				"INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents[sdk=iphonesimulator*]" = YES;
+				"INFOPLIST_KEY_UILaunchScreen_Generation[sdk=iphoneos*]" = YES;
+				"INFOPLIST_KEY_UILaunchScreen_Generation[sdk=iphonesimulator*]" = YES;
+				"INFOPLIST_KEY_UIStatusBarStyle[sdk=iphoneos*]" = UIStatusBarStyleDefault;
+				"INFOPLIST_KEY_UIStatusBarStyle[sdk=iphonesimulator*]" = UIStatusBarStyleDefault;
+				INFOPLIST_KEY_UISupportedInterfaceOrientations_iPad = "UIInterfaceOrientationPortrait UIInterfaceOrientationPortraitUpsideDown UIInterfaceOrientationLandscapeLeft UIInterfaceOrientationLandscapeRight";
+				INFOPLIST_KEY_UISupportedInterfaceOrientations_iPhone = "UIInterfaceOrientationPortrait UIInterfaceOrientationLandscapeLeft UIInterfaceOrientationLandscapeRight";
+				IPHONEOS_DEPLOYMENT_TARGET = 16.1;
+				LD_RUNPATH_SEARCH_PATHS = "@executable_path/Frameworks";
+				"LD_RUNPATH_SEARCH_PATHS[sdk=macosx*]" = "@executable_path/../Frameworks";
+				LLVM_LTO = YES;
+				MACOSX_DEPLOYMENT_TARGET = 13.0;
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = com.whispercppdemo.WhisperCppDemo;
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				SDKROOT = auto;
+				SUPPORTED_PLATFORMS = "iphoneos iphonesimulator macosx";
+				SWIFT_EMIT_LOC_STRINGS = YES;
+				SWIFT_OBJC_BRIDGING_HEADER = "whisper.cpp.swift/WhisperCppDemo-Bridging-Header.h";
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+			};
+			name = Release;
+		};
+/* End XCBuildConfiguration section */
+
+/* Begin XCConfigurationList section */
+		0AAC5D9229539CCF003032C3 /* Build configuration list for PBXProject "whisper.swiftui" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				0AAC5DBA29539CD0003032C3 /* Debug */,
+				0AAC5DBB29539CD0003032C3 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+		0AAC5DBC29539CD0003032C3 /* Build configuration list for PBXNativeTarget "whisper.swiftui" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				0AAC5DBD29539CD0003032C3 /* Debug */,
+				0AAC5DBE29539CD0003032C3 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+/* End XCConfigurationList section */
+	};
+	rootObject = 0AAC5D8F29539CCF003032C3 /* Project object */;
+}

examples/whisper.swiftui/whisper.swiftui.xcodeproj/project.xcworkspace/.gitignore

@@ -0,0 +1 @@
+contents.xcworkspacedata

examples/whisper.swiftui/whisper.swiftui.xcodeproj/project.xcworkspace/xcshareddata/IDEWorkspaceChecks.plist

@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+	<key>IDEDidComputeMac32BitWarning</key>
+	<true/>
+</dict>
+</plist>

examples/whisper.swiftui/whisper.swiftui.xcodeproj/xcshareddata/xcschemes/WhisperCppDemo.xcscheme

@@ -0,0 +1,100 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<Scheme
+   LastUpgradeVersion = "1410"
+   version = "1.3">
+   <BuildAction
+      parallelizeBuildables = "YES"
+      buildImplicitDependencies = "YES">
+      <BuildActionEntries>
+         <BuildActionEntry
+            buildForTesting = "YES"
+            buildForRunning = "YES"
+            buildForProfiling = "YES"
+            buildForArchiving = "YES"
+            buildForAnalyzing = "YES">
+            <BuildableReference
+               BuildableIdentifier = "primary"
+               BlueprintIdentifier = "0AAC5D9629539CCF003032C3"
+               BuildableName = "whisper.swiftui.app"
+               BlueprintName = "whisper.swiftui"
+               ReferencedContainer = "container:whisper.swiftui.xcodeproj">
+            </BuildableReference>
+         </BuildActionEntry>
+      </BuildActionEntries>
+   </BuildAction>
+   <TestAction
+      buildConfiguration = "Debug"
+      selectedDebuggerIdentifier = "Xcode.DebuggerFoundation.Debugger.LLDB"
+      selectedLauncherIdentifier = "Xcode.DebuggerFoundation.Launcher.LLDB"
+      shouldUseLaunchSchemeArgsEnv = "YES">
+      <Testables>
+         <TestableReference
+            skipped = "NO"
+            parallelizable = "YES">
+            <BuildableReference
+               BuildableIdentifier = "primary"
+               BlueprintIdentifier = "0AAC5DA729539CD0003032C3"
+               BuildableName = "whisper.swiftuiTests.xctest"
+               BlueprintName = "whisper.swiftuiTests"
+               ReferencedContainer = "container:whisper.swiftui.xcodeproj">
+            </BuildableReference>
+         </TestableReference>
+         <TestableReference
+            skipped = "NO"
+            parallelizable = "YES">
+            <BuildableReference
+               BuildableIdentifier = "primary"
+               BlueprintIdentifier = "0AAC5DB129539CD0003032C3"
+               BuildableName = "whisper.swiftuiUITests.xctest"
+               BlueprintName = "whisper.swiftuiUITests"
+               ReferencedContainer = "container:whisper.swiftui.xcodeproj">
+            </BuildableReference>
+         </TestableReference>
+      </Testables>
+   </TestAction>
+   <LaunchAction
+      buildConfiguration = "Release"
+      selectedDebuggerIdentifier = "Xcode.DebuggerFoundation.Debugger.LLDB"
+      selectedLauncherIdentifier = "Xcode.DebuggerFoundation.Launcher.LLDB"
+      launchStyle = "0"
+      useCustomWorkingDirectory = "NO"
+      ignoresPersistentStateOnLaunch = "NO"
+      debugDocumentVersioning = "YES"
+      debugServiceExtension = "internal"
+      allowLocationSimulation = "YES">
+      <BuildableProductRunnable
+         runnableDebuggingMode = "0">
+         <BuildableReference
+            BuildableIdentifier = "primary"
+            BlueprintIdentifier = "0AAC5D9629539CCF003032C3"
+            BuildableName = "whisper.swiftui.app"
+            BlueprintName = "whisper.swiftui"
+            ReferencedContainer = "container:whisper.swiftui.xcodeproj">
+         </BuildableReference>
+      </BuildableProductRunnable>
+   </LaunchAction>
+   <ProfileAction
+      buildConfiguration = "Release"
+      shouldUseLaunchSchemeArgsEnv = "YES"
+      savedToolIdentifier = ""
+      useCustomWorkingDirectory = "NO"
+      debugDocumentVersioning = "YES">
+      <BuildableProductRunnable
+         runnableDebuggingMode = "0">
+         <BuildableReference
+            BuildableIdentifier = "primary"
+            BlueprintIdentifier = "0AAC5D9629539CCF003032C3"
+            BuildableName = "whisper.swiftui.app"
+            BlueprintName = "whisper.swiftui"
+            ReferencedContainer = "container:whisper.swiftui.xcodeproj">
+         </BuildableReference>
+      </BuildableProductRunnable>
+   </ProfileAction>
+   <AnalyzeAction
+      buildConfiguration = "Debug">
+   </AnalyzeAction>
+   <ArchiveAction
+      buildConfiguration = "Release"
+      revealArchiveInOrganizer = "YES">
+   </ArchiveAction>
+</Scheme>

examples/whisper.wasm/CMakeLists.txt

@@ -8,6 +8,8 @@ add_executable(${TARGET}
     emscripten.cpp
     )
 
+include(DefaultTargetOptions)
+
 target_link_libraries(${TARGET} PRIVATE
     whisper
     )
@@ -30,8 +32,8 @@ set_target_properties(${TARGET} PROPERTIES LINK_FLAGS " \
     --bind \
     -s USE_PTHREADS=1 \
     -s PTHREAD_POOL_SIZE=8 \
-    -s INITIAL_MEMORY=1024MB \
-    -s TOTAL_MEMORY=1024MB \
+    -s INITIAL_MEMORY=1500MB \
+    -s TOTAL_MEMORY=1500MB \
     -s FORCE_FILESYSTEM=1 \
     -s EXPORTED_RUNTIME_METHODS=\"['print', 'printErr', 'ccall', 'cwrap']\" \
     ${EXTRA_FLAGS} \

examples/whisper.wasm/emscripten.cpp

@@ -18,7 +18,7 @@ EMSCRIPTEN_BINDINGS(whisper) {
 
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init(path_model.c_str());
+                g_contexts[i] = whisper_init_from_file(path_model.c_str());
                 if (g_contexts[i] != nullptr) {
                     return i + 1;
                 } else {

examples/whisper.wasm/index-tmpl.html

@@ -46,10 +46,12 @@
 
             <div id="model">
                 Whisper model: <span id="model-whisper-status"></span>
-                <button id="fetch-whisper-tiny-en" onclick="loadWhisper('tiny.en')">tiny.en (75 MB)</button>
-                <button id="fetch-whisper-tiny"    onclick="loadWhisper('tiny')">tiny (75 MB)</button>
-                <button id="fetch-whisper-base-en" onclick="loadWhisper('base.en')">base.en (142 MB)</button>
-                <button id="fetch-whisper-base"    onclick="loadWhisper('base')">base (142 MB)</button>
+                <button id="fetch-whisper-tiny-en"  onclick="loadWhisper('tiny.en')">tiny.en (75 MB)</button>
+                <button id="fetch-whisper-tiny"     onclick="loadWhisper('tiny')">tiny (75 MB)</button>
+                <button id="fetch-whisper-base-en"  onclick="loadWhisper('base.en')">base.en (142 MB)</button>
+                <button id="fetch-whisper-base"     onclick="loadWhisper('base')">base (142 MB)</button>
+                <button id="fetch-whisper-small-en" onclick="loadWhisper('small.en')">small.en (466 MB)</button>
+                <button id="fetch-whisper-small"    onclick="loadWhisper('small')">small (466 MB)</button>
                 <span id="fetch-whisper-progress"></span>
 
                 <input type="file" id="whisper-file" name="file" onchange="loadFile(event, 'whisper.bin')" />
@@ -60,8 +62,8 @@
             <!-- radio button to select between file upload or microphone -->
             <div id="input">
                 Input:
-                <input type="radio" id="file" name="input" value="file" checked="checked" onchange="changeInput('file')" /> File
-                <input type="radio" id="mic" name="input" value="mic" onchange="changeInput('mic')" /> Microphone
+                <input type="radio" id="file" name="input" value="file" checked="checked" onchange="changeInput('file')" /> <label for="file">File</label>
+                <input type="radio" id="mic" name="input" value="mic" onchange="changeInput('mic')" /> <label for="mic">Microphone</label>
             </div>
 
             <br>
@@ -284,27 +286,33 @@
                 }
                 reader.readAsArrayBuffer(file);
 
-                document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
-                document.getElementById('fetch-whisper-base-en').style.display = 'none';
-                document.getElementById('fetch-whisper-tiny'   ).style.display = 'none';
-                document.getElementById('fetch-whisper-base'   ).style.display = 'none';
-                document.getElementById('whisper-file'         ).style.display = 'none';
-                document.getElementById('model-whisper-status' ).innerHTML = 'loaded model: ' + file.name;
+                document.getElementById('fetch-whisper-tiny-en' ).style.display = 'none';
+                document.getElementById('fetch-whisper-base-en' ).style.display = 'none';
+                document.getElementById('fetch-whisper-small-en').style.display = 'none';
+                document.getElementById('fetch-whisper-tiny'    ).style.display = 'none';
+                document.getElementById('fetch-whisper-base'    ).style.display = 'none';
+                document.getElementById('fetch-whisper-small'   ).style.display = 'none';
+                document.getElementById('whisper-file'          ).style.display = 'none';
+                document.getElementById('model-whisper-status'  ).innerHTML = 'loaded model: ' + file.name;
             }
 
             function loadWhisper(model) {
                 let urls = {
-                    'tiny.en': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
-                    'tiny':    'https://whisper.ggerganov.com/ggml-model-whisper-tiny.bin',
-                    'base.en': 'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
-                    'base':    'https://whisper.ggerganov.com/ggml-model-whisper-base.bin',
+                    'tiny.en':  'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
+                    'tiny':     'https://whisper.ggerganov.com/ggml-model-whisper-tiny.bin',
+                    'base.en':  'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
+                    'base':     'https://whisper.ggerganov.com/ggml-model-whisper-base.bin',
+                    'small.en': 'https://whisper.ggerganov.com/ggml-model-whisper-small.en.bin',
+                    'small':    'https://whisper.ggerganov.com/ggml-model-whisper-small.bin',
                 };
 
                 let sizes = {
-                    'tiny.en': 75,
-                    'tiny':    75,
-                    'base.en': 142,
-                    'base':    142,
+                    'tiny.en':  75,
+                    'tiny':     75,
+                    'base.en':  142,
+                    'base':     142,
+                    'small.en': 466,
+                    'small':    466,
                 };
 
                 let url     = urls[model];
@@ -313,12 +321,14 @@
 
                 model_whisper = model;
 
-                document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
-                document.getElementById('fetch-whisper-base-en').style.display = 'none';
-                document.getElementById('fetch-whisper-tiny'   ).style.display = 'none';
-                document.getElementById('fetch-whisper-base'   ).style.display = 'none';
-                document.getElementById('whisper-file'         ).style.display = 'none';
-                document.getElementById('model-whisper-status' ).innerHTML = 'loading model: ' + model;
+                document.getElementById('fetch-whisper-tiny-en' ).style.display = 'none';
+                document.getElementById('fetch-whisper-base-en' ).style.display = 'none';
+                document.getElementById('fetch-whisper-small-en').style.display = 'none';
+                document.getElementById('fetch-whisper-tiny'    ).style.display = 'none';
+                document.getElementById('fetch-whisper-base'    ).style.display = 'none';
+                document.getElementById('fetch-whisper-small'   ).style.display = 'none';
+                document.getElementById('whisper-file'          ).style.display = 'none';
+                document.getElementById('model-whisper-status'  ).innerHTML = 'loading model: ' + model;
 
                 cbProgress = function(p) {
                     let el = document.getElementById('fetch-whisper-progress');
@@ -327,12 +337,14 @@
 
                 cbCancel = function() {
                     var el;
-                    el = document.getElementById('fetch-whisper-tiny-en'); if (el) el.style.display = 'inline-block';
-                    el = document.getElementById('fetch-whisper-base-en'); if (el) el.style.display = 'inline-block';
-                    el = document.getElementById('fetch-whisper-tiny'   ); if (el) el.style.display = 'inline-block';
-                    el = document.getElementById('fetch-whisper-base'   ); if (el) el.style.display = 'inline-block';
-                    el = document.getElementById('whisper-file'         ); if (el) el.style.display = 'inline-block';
-                    el = document.getElementById('model-whisper-status' ); if (el) el.innerHTML = '';
+                    el = document.getElementById('fetch-whisper-tiny-en' ); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('fetch-whisper-base-en' ); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('fetch-whisper-small-en'); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('fetch-whisper-tiny'    ); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('fetch-whisper-base'    ); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('fetch-whisper-small'   ); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('whisper-file'          ); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('model-whisper-status'  ); if (el) el.innerHTML = '';
                 };
 
                 loadRemote(url, dst, size_mb, cbProgress, storeFS, cbCancel, printTextarea);

extra/bench-all.sh

@@ -12,6 +12,18 @@ fi
 
 models=( "tiny" "base" "small" "medium" "large" )
 
+printf "\n"
+printf "Running memcpy benchmark with 1 thread\n"
+printf "\n"
+
+./bench -w 1 -t 1 2>&1
+
+printf "\n"
+printf "Running ggml_mul_mat benchmark with $n_threads threads\n"
+printf "\n"
+
+./bench -w 2 -t $n_threads 2>&1
+
 printf "\n"
 printf "Running benchmark for all models\n"
 printf "This can take a while!\n"
@@ -56,4 +68,3 @@ for model in "${models[@]}"; do
 
     printf "| <todo> | <todo> | $config | $model | $n_threads | $load_time | $encode_time | $commit |\n"
 done
-