release : v1.5.2

* wchess: whisper assisted chess

* wchess: fix allowed moves in check

* wchess: touchstart, touchend events

* wchess: css, disabled button

* wchess : html touches

* wchess : minor fixes and code style

* wchess : bump encoder context to 1280

* wchess : index.html

* wchess : fix CI warnings

* wchess : add array header

* wchess : build static library

* wchess : display grammar

* wchess : update UX

* wchess : add comment

* wchess : add README

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

.github/workflows/build.yml

@@ -25,6 +25,7 @@ jobs:
           docker run --platform ${{ matrix.arch }} --rm \
             -v ${{ github.workspace }}:/workspace \
             -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
+            set -e
             apt update
             apt install -y build-essential libsdl2-dev
             make
@@ -86,6 +87,7 @@ jobs:
           docker run --platform ${{ matrix.arch }} --rm \
             -v ${{ github.workspace }}:/workspace \
             -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
+            set -e
             apt update
             apt install -y build-essential cmake libsdl2-dev
             cmake . -DWHISPER_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }}
@@ -113,8 +115,10 @@ jobs:
           docker run --platform ${{ matrix.arch }} --rm \
             -v ${{ github.workspace }}:/workspace \
             -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
+            set -e
             apt update
-            apt install -y build-essential cmake libsdl2-dev
+            apt install -y clang
+            apt install -y clang build-essential cmake libsdl2-dev
             cmake . -DWHISPER_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang
             make
             ctest -L gh --output-on-failure'
@@ -140,6 +144,7 @@ jobs:
           docker run --platform ${{ matrix.arch }} --rm \
             -v ${{ github.workspace }}:/workspace \
             -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
+            set -e
             apt update
             apt install -y build-essential cmake
             cmake . -DCMAKE_BUILD_TYPE=Debug -DWHISPER_SANITIZE_${{ matrix.sanitizer }}=ON
@@ -217,10 +222,10 @@ jobs:
         sdl2: [ON]
         include:
           - arch: Win32
-            obzip: https://github.com/OpenMathLib/OpenBLAS/releases/download/v0.3.24/OpenBLAS-0.3.24-x86.zip
+            obzip: https://github.com/OpenMathLib/OpenBLAS/releases/download/v0.3.25/OpenBLAS-0.3.25-x86.zip
             s2arc: x86
           - arch: x64
-            obzip: https://github.com/OpenMathLib/OpenBLAS/releases/download/v0.3.24/OpenBLAS-0.3.24-x64.zip
+            obzip: https://github.com/OpenMathLib/OpenBLAS/releases/download/v0.3.25/OpenBLAS-0.3.25-x64.zip
             s2arc: x64
           - sdl2: ON
             s2ver: 2.26.0
@@ -285,6 +290,7 @@ jobs:
         arch: [x64]
         cublas: [ON]
         sdl2: [ON]
+        cuda-toolkit: [12.2.0, 11.8.0]
         include:
           - arch: x64
             s2arc: x64
@@ -300,7 +306,9 @@ jobs:
 
       - name: Install CUDA Toolkit
         id: cuda-toolkit
-        uses: Jimver/cuda-toolkit@v0.2.10
+        uses: Jimver/cuda-toolkit@v0.2.11
+        with:
+          cuda: '${{ matrix.cuda-toolkit }}'
 
       - name: Fetch SDL2 and set SDL2_DIR
         if: matrix.sdl2 == 'ON'
@@ -315,10 +323,17 @@ jobs:
           -DCMAKE_BUILD_TYPE=${{ matrix.build }}
           -DWHISPER_CUBLAS=1
 
-      - name: Build
+      - name: Build ${{ matrix.cuda-toolkit }}
         run: |
           cd ./build
-          msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
+          cmake --build . --config ${{ matrix.build }}
+
+      - name: Copy CUDA DLLs
+        run: >
+          Copy-Item -PassThru
+          -Path "${{ steps.cuda-toolkit.outputs.CUDA_PATH }}/bin/*.dll"
+          -Include cudart64_*,cublas64_*,cublasLt64_*
+          -Destination build/bin/${{ matrix.build }}
 
       - name: Copy SDL2.dll
         if: matrix.sdl2 == 'ON'
@@ -328,7 +343,7 @@ jobs:
         if: matrix.sdl2 == 'ON'
         uses: actions/upload-artifact@v1
         with:
-          name: whisper-cublas-bin-${{ matrix.arch }}
+          name: whisper-cublas-${{ matrix.cuda-toolkit }}-bin-${{ matrix.arch }}
           path: build/bin/${{ matrix.build }}
 
   emscripten:
@@ -396,6 +411,32 @@ jobs:
           cd examples/whisper.android
           ./gradlew assembleRelease --no-daemon
 
+  android_java:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Clone
+        uses: actions/checkout@v3
+
+      - name: set up JDK 11
+        uses: actions/setup-java@v3
+        with:
+          java-version: '11'
+          distribution: 'temurin'
+          cache: gradle
+
+      - name: Setup Android SDK
+        uses: android-actions/setup-android@v2
+        with:
+          api-level: 30
+          build-tools-version: 30.0.3
+
+      - name: Build
+        run: |
+          cd examples/whisper.android.java
+          chmod +x ./gradlew 
+          ./gradlew assembleRelease
+
   java:
     needs: [ 'windows' ]
     runs-on: windows-latest

.gitignore

@@ -8,6 +8,7 @@
 .DS_Store
 
 build/
+build-coreml/
 build-em/
 build-debug/
 build-release/
@@ -30,6 +31,7 @@ build-sanitize-thread/
 /talk-llama
 /bench
 /quantize
+/server
 /lsp
 
 arm_neon.h
@@ -53,3 +55,7 @@ bindings/java/.idea/
 .idea/
 
 benchmark_results.csv
+cmake-build-debug/
+.cxx/
+.gradle/
+local.properties

CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required (VERSION 3.5)
 
-project(whisper.cpp VERSION 1.4.3)
+project(whisper.cpp VERSION 1.5.2)
 
 # Add path to modules
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
@@ -533,7 +533,7 @@ target_compile_definitions(${TARGET} PUBLIC
     ${WHISPER_EXTRA_FLAGS}
     )
 
-set_target_properties(${TARGET} PROPERTIES PUBLIC_HEADER "whisper.h")
+set_target_properties(${TARGET} PROPERTIES PUBLIC_HEADER "ggml.h;whisper.h")
 
 include(GNUInstallDirs)
 

Makefile

@@ -1,4 +1,4 @@
-default: main bench quantize
+default: main bench quantize server
 
 ifndef UNAME_S
 UNAME_S := $(shell uname -s)
@@ -307,7 +307,7 @@ ggml-backend.o: ggml-backend.c ggml.h ggml-backend.h
 ggml-quants.o: ggml-quants.c ggml.h ggml-quants.h
 	$(CC)  $(CFLAGS)   -c $< -o $@
 
-WHISPER_OBJ += ggml-alloc.o ggml-backend.o ggml-quants.o
+WHISPER_OBJ += ggml.o ggml-alloc.o ggml-backend.o ggml-quants.o
 
 whisper.o: whisper.cpp whisper.h ggml.h ggml-cuda.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
@@ -331,14 +331,14 @@ ggml-metal.o: ggml-metal.m ggml-metal.h
 WHISPER_OBJ += ggml-metal.o
 endif
 
-libwhisper.a: ggml.o $(WHISPER_OBJ)
-	$(AR) rcs libwhisper.a ggml.o $(WHISPER_OBJ)
+libwhisper.a: $(WHISPER_OBJ)
+	$(AR) rcs libwhisper.a $(WHISPER_OBJ)
 
-libwhisper.so: ggml.o $(WHISPER_OBJ)
-	$(CXX) $(CXXFLAGS) -shared -o libwhisper.so ggml.o $(WHISPER_OBJ) $(LDFLAGS)
+libwhisper.so: $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) -shared -o libwhisper.so $(WHISPER_OBJ) $(LDFLAGS)
 
 clean:
-	rm -f *.o main stream command talk talk-llama bench quantize lsp libwhisper.a libwhisper.so
+	rm -f *.o main stream command talk talk-llama bench quantize server lsp libwhisper.a libwhisper.so
 
 #
 # Examples
@@ -349,30 +349,33 @@ CC_SDL=`sdl2-config --cflags --libs`
 SRC_COMMON     = examples/common.cpp examples/common-ggml.cpp
 SRC_COMMON_SDL = examples/common-sdl.cpp
 
-main: examples/main/main.cpp $(SRC_COMMON) ggml.o $(WHISPER_OBJ)
-	$(CXX) $(CXXFLAGS) examples/main/main.cpp $(SRC_COMMON) ggml.o $(WHISPER_OBJ) -o main $(LDFLAGS)
+main: examples/main/main.cpp $(SRC_COMMON) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/main/main.cpp $(SRC_COMMON) $(WHISPER_OBJ) -o main $(LDFLAGS)
 	./main -h
 
-bench: examples/bench/bench.cpp ggml.o $(WHISPER_OBJ)
-	$(CXX) $(CXXFLAGS) examples/bench/bench.cpp ggml.o $(WHISPER_OBJ) -o bench $(LDFLAGS)
+bench: examples/bench/bench.cpp $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/bench/bench.cpp $(WHISPER_OBJ) -o bench $(LDFLAGS)
 
-quantize: examples/quantize/quantize.cpp ggml.o $(WHISPER_OBJ) $(SRC_COMMON)
-	$(CXX) $(CXXFLAGS) examples/quantize/quantize.cpp $(SRC_COMMON) ggml.o $(WHISPER_OBJ) -o quantize $(LDFLAGS)
+quantize: examples/quantize/quantize.cpp $(WHISPER_OBJ) $(SRC_COMMON)
+	$(CXX) $(CXXFLAGS) examples/quantize/quantize.cpp $(SRC_COMMON) $(WHISPER_OBJ) -o quantize $(LDFLAGS)
 
-stream: examples/stream/stream.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ)
-	$(CXX) $(CXXFLAGS) examples/stream/stream.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ) -o stream $(CC_SDL) $(LDFLAGS)
+server: examples/server/server.cpp $(SRC_COMMON) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/server/server.cpp $(SRC_COMMON) $(WHISPER_OBJ) -o server $(LDFLAGS)
 
-command: examples/command/command.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ)
-	$(CXX) $(CXXFLAGS) examples/command/command.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ) -o command $(CC_SDL) $(LDFLAGS)
+stream: examples/stream/stream.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/stream/stream.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o stream $(CC_SDL) $(LDFLAGS)
 
-lsp: examples/lsp/lsp.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ)
-	$(CXX) $(CXXFLAGS) examples/lsp/lsp.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ) -o lsp $(CC_SDL) $(LDFLAGS)
+command: examples/command/command.cpp examples/grammar-parser.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/command/command.cpp examples/grammar-parser.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o command $(CC_SDL) $(LDFLAGS)
 
-talk: examples/talk/talk.cpp examples/talk/gpt-2.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ)
-	$(CXX) $(CXXFLAGS) examples/talk/talk.cpp examples/talk/gpt-2.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ) -o talk $(CC_SDL) $(LDFLAGS)
+lsp: examples/lsp/lsp.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/lsp/lsp.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o lsp $(CC_SDL) $(LDFLAGS)
 
-talk-llama: examples/talk-llama/talk-llama.cpp examples/talk-llama/llama.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ)
-	$(CXX) $(CXXFLAGS) examples/talk-llama/talk-llama.cpp examples/talk-llama/llama.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o $(WHISPER_OBJ) -o talk-llama $(CC_SDL) $(LDFLAGS)
+talk: examples/talk/talk.cpp examples/talk/gpt-2.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/talk/talk.cpp examples/talk/gpt-2.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o talk $(CC_SDL) $(LDFLAGS)
+
+talk-llama: examples/talk-llama/talk-llama.cpp examples/talk-llama/llama.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/talk-llama/talk-llama.cpp examples/talk-llama/llama.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o talk-llama $(CC_SDL) $(LDFLAGS)
 
 #
 # Audio samples
@@ -418,9 +421,9 @@ samples:
 .PHONY: medium
 .PHONY: large-v1
 .PHONY: large-v2
-.PHONY: large
+.PHONY: large-v3
 
-tiny.en tiny base.en base small.en small medium.en medium large-v1 large-v2 large: main
+tiny.en tiny base.en base small.en small medium.en medium large-v1 large-v2 large-v3: main
 	bash ./models/download-ggml-model.sh $@
 	@echo ""
 	@echo "==============================================="

Package.swift

@@ -2,33 +2,14 @@
 
 import PackageDescription
 
-#if arch(arm) || arch(arm64)
-let platforms: [SupportedPlatform]? = [
-    .macOS(.v12),
-    .iOS(.v14),
-    .watchOS(.v4),
-    .tvOS(.v14)
-]
-let exclude: [String] = []
-let resources: [Resource] = [
-    .process("ggml-metal.metal")
-]
-let additionalSources: [String] = ["ggml-metal.m"]
-let additionalSettings: [CSetting] = [
-    .unsafeFlags(["-fno-objc-arc"]),
-    .define("GGML_USE_METAL")
-]
-#else
-let platforms: [SupportedPlatform]? = nil
-let exclude: [String] = ["ggml-metal.metal"]
-let resources: [Resource] = []
-let additionalSources: [String] = []
-let additionalSettings: [CSetting] = []
-#endif
-
 let package = Package(
     name: "whisper",
-    platforms: platforms,
+    platforms: [
+        .macOS(.v12),
+        .iOS(.v14),
+        .watchOS(.v4),
+        .tvOS(.v14)
+    ],
     products: [
         .library(name: "whisper", targets: ["whisper"]),
     ],
@@ -36,7 +17,7 @@ let package = Package(
         .target(
             name: "whisper",
             path: ".",
-            exclude: exclude + [
+            exclude: [
                "bindings",
                "cmake",
                "coreml",
@@ -55,19 +36,22 @@ let package = Package(
                 "whisper.cpp",
                 "ggml-alloc.c",
                 "ggml-backend.c",
-                "ggml-quants.c"
-            ] + additionalSources,
-            resources: resources,
+                "ggml-quants.c",
+                "ggml-metal.m"
+            ],
+            resources: [.process("ggml-metal.metal")],
             publicHeadersPath: "spm-headers",
             cSettings: [
                 .unsafeFlags(["-Wno-shorten-64-to-32", "-O3", "-DNDEBUG"]),
-                .define("GGML_USE_ACCELERATE")
+                .define("GGML_USE_ACCELERATE"),
+                .unsafeFlags(["-fno-objc-arc"]),
+                .define("GGML_USE_METAL")
                 // NOTE: NEW_LAPACK will required iOS version 16.4+
                 // We should consider add this in the future when we drop support for iOS 14
                 // (ref: ref: https://developer.apple.com/documentation/accelerate/1513264-cblas_sgemm?language=objc)
                 // .define("ACCELERATE_NEW_LAPACK"),
                 // .define("ACCELERATE_LAPACK_ILP64")
-            ] + additionalSettings,
+            ],
             linkerSettings: [
                 .linkedFramework("Accelerate")
             ]

README.md

@@ -6,7 +6,7 @@
 [![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/)
 
-Beta: [v1.4.3](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.4.3) / Stable: [v1.2.1](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.2.1) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
+Stable: [v1.5.2](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.5.2) / [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:
 
@@ -16,12 +16,10 @@ High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisp
 - VSX intrinsics support for POWER architectures
 - Mixed F16 / F32 precision
 - [4-bit and 5-bit integer quantization support](https://github.com/ggerganov/whisper.cpp#quantization)
-- Low memory usage (Flash Attention)
 - Zero memory allocations at runtime
 - Support for CPU-only inference
-- [Partial GPU support for NVIDIA via cuBLAS](https://github.com/ggerganov/whisper.cpp#nvidia-gpu-support-via-cublas)
+- [Efficient GPU support for NVIDIA](https://github.com/ggerganov/whisper.cpp#nvidia-gpu-support-via-cublas)
 - [Partial OpenCL GPU support via CLBlast](https://github.com/ggerganov/whisper.cpp#opencl-gpu-support-via-clblast)
-- [BLAS CPU support via OpenBLAS](https://github.com/ggerganov/whisper.cpp#blas-cpu-support-via-openblas)
 - [OpenVINO Support](https://github.com/ggerganov/whisper.cpp#openvino-support)
 - [C-style API](https://github.com/ggerganov/whisper.cpp/blob/master/whisper.h)
 
@@ -36,10 +34,8 @@ Supported platforms:
 - [x] Windows ([MSVC](https://github.com/ggerganov/whisper.cpp/blob/master/.github/workflows/build.yml#L117-L144) and [MinGW](https://github.com/ggerganov/whisper.cpp/issues/168)]
 - [x] [Raspberry Pi](https://github.com/ggerganov/whisper.cpp/discussions/166)
 
-The entire implementation of the model is contained in 2 source files:
-
-- Tensor operations: [ggml.h](ggml.h) / [ggml.c](ggml.c)
-- Transformer inference: [whisper.h](whisper.h) / [whisper.cpp](whisper.cpp)
+The entire high-level implementation of the model is contained in [whisper.h](whisper.h) and [whisper.cpp](whisper.cpp).
+The rest of the code is part of the [ggml](https://github.com/ggerganov/ggml) machine learning library.
 
 Having such a lightweight implementation of the model allows to easily integrate it in different platforms and applications.
 As an example, here is a video of running the model on an iPhone 13 device - fully offline, on-device: [whisper.objc](examples/whisper.objc)
@@ -114,8 +110,8 @@ options:
   -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
   -sow,      --split-on-word     [false  ] split on word rather than on token
-  -bo N,     --best-of N         [2      ] number of best candidates to keep
-  -bs N,     --beam-size N       [-1     ] beam size for beam search
+  -bo N,     --best-of N         [5      ] number of best candidates to keep
+  -bs N,     --beam-size N       [5      ] 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
@@ -132,6 +128,7 @@ options:
   -fp,       --font-path         [/System/Library/Fonts/Supplemental/Courier New Bold.ttf] path to a monospace font for karaoke video
   -ocsv,     --output-csv        [false  ] output result in a CSV file
   -oj,       --output-json       [false  ] output result in a JSON file
+  -ojf,      --output-json-full  [false  ] include more information in the JSON 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
@@ -143,7 +140,8 @@ options:
   -m FNAME,  --model FNAME       [models/ggml-base.en.bin] model path
   -f FNAME,  --file FNAME        [       ] input WAV file path
   -oved D,   --ov-e-device DNAME [CPU    ] the OpenVINO device used for encode inference
-  -ls,       --log-score         [false  ] log best decoder scores of token
+  -ls,       --log-score         [false  ] log best decoder scores of tokens
+  -ng,       --no-gpu            [false  ] disable GPU
 
 
 bash ./models/download-ggml-model.sh base.en
@@ -235,18 +233,18 @@ make medium.en
 make medium
 make large-v1
 make large-v2
-make large
+make large-v3
 ```
 
 ## Memory usage
 
-| Model  | Disk   | Mem     | SHA                                        |
-| ---    | ---    | ---     | ---                                        |
-| 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 | `ad82bf6a9043ceed055076d0fd39f5f186ff8062` |
+| Model  | Disk    | Mem      |
+| ---    | ---     | ---      |
+| tiny   |  75 MiB | ~273 MB |
+| base   | 142 MiB | ~388 MB |
+| small  | 466 MiB | ~852 MB |
+| medium | 1.5 GiB | ~2.1 GB |
+| large  | 2.9 GiB | ~3.9 GB |
 
 ## Quantization
 
@@ -400,12 +398,12 @@ This can result in significant speedup in encoder performance. Here are the inst
 
   The first time run on an OpenVINO device is slow, since the OpenVINO framework will compile the IR (Intermediate Representation) model to a device-specific 'blob'. This device-specific blob will get
   cached for the next run.
-  
+
 For more information about the Core ML implementation please refer to PR [#1037](https://github.com/ggerganov/whisper.cpp/pull/1037).
 
-## NVIDIA GPU support via cuBLAS
+## NVIDIA GPU support
 
-With NVIDIA cards the Encoder processing can to a large extent be offloaded to the GPU through cuBLAS.
+With NVIDIA cards the processing of the models is done efficiently on the GPU via cuBLAS and custom CUDA kernels.
 First, make sure you have installed `cuda`: https://developer.nvidia.com/cuda-downloads
 
 Now build `whisper.cpp` with cuBLAS support:
@@ -772,6 +770,7 @@ Some of the examples are even ported to run in the browser using WebAssembly. Ch
 | [bench](examples/bench) | [bench.wasm](examples/bench.wasm) | Benchmark the performance of Whisper on your machine |
 | [stream](examples/stream) | [stream.wasm](examples/stream.wasm) | Real-time transcription of raw microphone capture |
 | [command](examples/command) | [command.wasm](examples/command.wasm) | Basic voice assistant example for receiving voice commands from the mic |
+| [wchess](examples/wchess) | [wchess.wasm](examples/wchess) | Voice-controlled chess |
 | [talk](examples/talk) | [talk.wasm](examples/talk.wasm) | Talk with a GPT-2 bot |
 | [talk-llama](examples/talk-llama) | | Talk with a LLaMA bot |
 | [whisper.objc](examples/whisper.objc) | | iOS mobile application using whisper.cpp |
@@ -781,6 +780,7 @@ Some of the examples are even ported to run in the browser using WebAssembly. Ch
 | [generate-karaoke.sh](examples/generate-karaoke.sh) | | Helper script to easily [generate a karaoke video](https://youtu.be/uj7hVta4blM) of raw audio capture |
 | [livestream.sh](examples/livestream.sh) | | [Livestream audio transcription](https://github.com/ggerganov/whisper.cpp/issues/185) |
 | [yt-wsp.sh](examples/yt-wsp.sh) | | Download + transcribe and/or translate any VOD [(original)](https://gist.github.com/DaniruKun/96f763ec1a037cc92fe1a059b643b818) |
+| [server](examples/server) | | HTTP transcription server with OAI-like API |
 
 ## [Discussions](https://github.com/ggerganov/whisper.cpp/discussions)
 

bindings/go/Makefile

@@ -1,9 +1,26 @@
+ifndef UNAME_S
+UNAME_S := $(shell uname -s)
+endif
+
+ifndef UNAME_P
+UNAME_P := $(shell uname -p)
+endif
+
+ifndef UNAME_M
+UNAME_M := $(shell uname -m)
+endif
+
+GGML_METAL_PATH_RESOURCES := $(abspath ../..)
 BUILD_DIR := build
 MODELS_DIR := models
 EXAMPLES_DIR := $(wildcard examples/*)
 INCLUDE_PATH := $(abspath ../..)
 LIBRARY_PATH := $(abspath ../..)
 
+ifeq ($(UNAME_S),Darwin)
+	EXT_LDFLAGS := -framework Foundation -framework Metal -framework MetalKit
+endif
+
 all: clean whisper examples
 
 whisper: mkdir
@@ -11,8 +28,13 @@ whisper: mkdir
 	@${MAKE} -C ../.. libwhisper.a
 
 test: model-small whisper modtidy
+ifeq ($(UNAME_S),Darwin)
+	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} GGML_METAL_PATH_RESOURCES=${GGML_METAL_PATH_RESOURCES} go test -ldflags "-extldflags '$(EXT_LDFLAGS)'" -v .
+	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} GGML_METAL_PATH_RESOURCES=${GGML_METAL_PATH_RESOURCES} go test -ldflags "-extldflags '$(EXT_LDFLAGS)'" -v ./pkg/whisper/...
+else
 	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} go test -v .
 	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} go test -v ./pkg/whisper/...
+endif
 
 examples: $(EXAMPLES_DIR)
 
@@ -21,7 +43,11 @@ model-small: mkdir examples/go-model-download
 
 $(EXAMPLES_DIR): mkdir whisper modtidy
 	@echo Build example $(notdir $@)
+ifeq ($(UNAME_S),Darwin)
+	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} GGML_METAL_PATH_RESOURCES=${GGML_METAL_PATH_RESOURCES} go build ${BUILD_FLAGS} -ldflags "-extldflags '$(EXT_LDFLAGS)'" -o ${BUILD_DIR}/$(notdir $@) ./$@
+else
 	@C_INCLUDE_PATH=${INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} go build ${BUILD_FLAGS} -o ${BUILD_DIR}/$(notdir $@) ./$@
+endif
 
 mkdir:
 	@echo Mkdir ${BUILD_DIR}

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

@@ -24,7 +24,7 @@ const (
 
 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-v2", "ggml-large"}
+	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-v2", "ggml-large-v3"}
 )
 
 var (

bindings/java/build.gradle

@@ -9,6 +9,7 @@ archivesBaseName = 'whispercpp'
 group = 'io.github.ggerganov'
 version = '1.4.0'
 
+
 sourceCompatibility = 1.8
 targetCompatibility = 1.8
 

bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCpp.java

@@ -2,6 +2,7 @@ package io.github.ggerganov.whispercpp;
 
 import com.sun.jna.Native;
 import com.sun.jna.Pointer;
+import io.github.ggerganov.whispercpp.bean.WhisperSegment;
 import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 import io.github.ggerganov.whispercpp.params.WhisperFullParams;
 import io.github.ggerganov.whispercpp.params.WhisperSamplingStrategy;
@@ -9,6 +10,8 @@ import io.github.ggerganov.whispercpp.params.WhisperSamplingStrategy;
 import java.io.File;
 import java.io.FileNotFoundException;
 import java.io.IOException;
+import java.util.ArrayList;
+import java.util.List;
 
 /**
  * Before calling most methods, you must call `initContext(modelPath)` to initialise the `ctx` Pointer.
@@ -160,6 +163,28 @@ public class WhisperCpp implements AutoCloseable {
 
         return str.toString().trim();
     }
+    public List<WhisperSegment> fullTranscribeWithTime(WhisperFullParams whisperParams, float[] audioData) throws IOException {
+        if (ctx == null) {
+            throw new IllegalStateException("Model not initialised");
+        }
+
+        if (lib.whisper_full(ctx, whisperParams, audioData, audioData.length) != 0) {
+            throw new IOException("Failed to process audio");
+        }
+
+        int nSegments = lib.whisper_full_n_segments(ctx);
+        List<WhisperSegment> segments= new ArrayList<>(nSegments);
+
+
+        for (int i = 0; i < nSegments; i++) {
+            long t0 = lib.whisper_full_get_segment_t0(ctx, i);
+            String text = lib.whisper_full_get_segment_text(ctx, i);
+            long t1 = lib.whisper_full_get_segment_t1(ctx, i);
+            segments.add(new WhisperSegment(t0,t1,text));
+        }
+
+        return segments;
+    }
 
 //    public int getTextSegmentCount(Pointer ctx) {
 //        return lib.whisper_full_n_segments(ctx);

bindings/java/src/main/java/io/github/ggerganov/whispercpp/bean/WhisperSegment.java

@@ -0,0 +1,47 @@
+package io.github.ggerganov.whispercpp.bean;
+
+/**
+ * Created by litonglinux@qq.com on 10/21/2023_7:48 AM
+ */
+public class WhisperSegment {
+  private long start, end;
+  private String sentence;
+
+  public WhisperSegment() {
+  }
+
+  public WhisperSegment(long start, long end, String sentence) {
+    this.start = start;
+    this.end = end;
+    this.sentence = sentence;
+  }
+
+  public long getStart() {
+    return start;
+  }
+
+  public long getEnd() {
+    return end;
+  }
+
+  public String getSentence() {
+    return sentence;
+  }
+
+  public void setStart(long start) {
+    this.start = start;
+  }
+
+  public void setEnd(long end) {
+    this.end = end;
+  }
+
+  public void setSentence(String sentence) {
+    this.sentence = sentence;
+  }
+
+  @Override
+  public String toString() {
+    return "[" + start + " --> " + end + "]:" + sentence;
+  }
+}

bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperFullParams.java

@@ -58,6 +58,9 @@ public class WhisperFullParams extends Structure {
         no_context = enable ? CBool.FALSE : CBool.TRUE;
     }
 
+    /** Generate timestamps or not? */
+    public CBool no_timestamps;
+
     /** Flag to force single segment output (useful for streaming). (default = false) */
     public CBool single_segment;
 
@@ -304,10 +307,16 @@ public class WhisperFullParams extends Structure {
         logits_filter_callback = CallbackReference.getFunctionPointer(callback);
     }
 
+    /** Grammar stuff */
+    public Pointer grammar_rules;
+    public long n_grammar_rules;
+    public long i_start_rule;
+    public float grammar_penalty;
+
     @Override
     protected List<String> getFieldOrder() {
         return Arrays.asList("strategy", "n_threads", "n_max_text_ctx", "offset_ms", "duration_ms", "translate",
-                "no_context", "single_segment",
+                "no_context", "single_segment", "no_timestamps",
                 "print_special", "print_progress", "print_realtime", "print_timestamps",  "token_timestamps",
                 "thold_pt", "thold_ptsum", "max_len", "split_on_word", "max_tokens", "speed_up", "audio_ctx",
                 "tdrz_enable", "initial_prompt", "prompt_tokens", "prompt_n_tokens", "language", "detect_language",
@@ -316,6 +325,7 @@ public class WhisperFullParams extends Structure {
                 "new_segment_callback", "new_segment_callback_user_data",
                 "progress_callback", "progress_callback_user_data",
                 "encoder_begin_callback", "encoder_begin_callback_user_data",
-                "logits_filter_callback", "logits_filter_callback_user_data");
+                "logits_filter_callback", "logits_filter_callback_user_data",
+                "grammar_rules", "n_grammar_rules", "i_start_rule", "grammar_penalty");
     }
 }

bindings/java/src/test/java/io/github/ggerganov/whispercpp/WhisperCppTest.java

@@ -2,6 +2,7 @@ package io.github.ggerganov.whispercpp;
 
 import static org.junit.jupiter.api.Assertions.*;
 
+import io.github.ggerganov.whispercpp.bean.WhisperSegment;
 import io.github.ggerganov.whispercpp.params.CBool;
 import io.github.ggerganov.whispercpp.params.WhisperFullParams;
 import io.github.ggerganov.whispercpp.params.WhisperSamplingStrategy;
@@ -11,6 +12,7 @@ import javax.sound.sampled.AudioInputStream;
 import javax.sound.sampled.AudioSystem;
 import java.io.File;
 import java.io.FileNotFoundException;
+import java.util.List;
 
 class WhisperCppTest {
     private static WhisperCpp whisper = new WhisperCpp();
@@ -20,11 +22,12 @@ class WhisperCppTest {
     static void init() throws FileNotFoundException {
         // By default, models are loaded from ~/.cache/whisper/ and are usually named "ggml-${name}.bin"
         // or you can provide the absolute path to the model file.
+        //String modelName = "../../models/ggml-tiny.bin";
         String modelName = "../../models/ggml-tiny.en.bin";
         try {
             whisper.initContext(modelName);
-//            whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
-//            whisper.getJavaDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
+            //whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
+            //whisper.getJavaDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
             modelInitialised = true;
         } catch (FileNotFoundException ex) {
             System.out.println("Model " + modelName + " not found");
@@ -42,7 +45,7 @@ class WhisperCppTest {
         assertEquals(16384, params.n_max_text_ctx);
         assertFalse(params.translate);
         assertEquals(0.01f, params.thold_pt);
-        assertEquals(2, params.beam_search.beam_size);
+        assertEquals(5, params.beam_search.beam_size);
         assertEquals(-1.0f, params.beam_search.patience);
     }
 
@@ -55,7 +58,7 @@ class WhisperCppTest {
         assertEquals(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY.ordinal(), params.strategy);
         assertNotEquals(0, params.n_threads);
         assertEquals(16384, params.n_max_text_ctx);
-        assertEquals(2, params.greedy.best_of);
+        assertEquals(5, params.greedy.best_of);
     }
 
     @Test
@@ -72,11 +75,11 @@ class WhisperCppTest {
         byte[] b = new byte[audioInputStream.available()];
         float[] floats = new float[b.length / 2];
 
-//        WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
+        //WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
         WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
         params.setProgressCallback((ctx, state, progress, user_data) -> System.out.println("progress: " + progress));
         params.print_progress = CBool.FALSE;
-//        params.initial_prompt = "and so my fellow Americans um, like";
+        //params.initial_prompt = "and so my fellow Americans um, like";
 
 
         try {
@@ -99,4 +102,43 @@ class WhisperCppTest {
             audioInputStream.close();
         }
     }
+
+    @Test
+    void testFullTranscribeWithTime() throws Exception {
+        if (!modelInitialised) {
+            System.out.println("Model not initialised, skipping test");
+            return;
+        }
+
+        // Given
+        File file = new File(System.getProperty("user.dir"), "../../samples/jfk.wav");
+        AudioInputStream audioInputStream = AudioSystem.getAudioInputStream(file);
+
+        byte[] b = new byte[audioInputStream.available()];
+        float[] floats = new float[b.length / 2];
+
+        //WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
+        WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
+        params.setProgressCallback((ctx, state, progress, user_data) -> System.out.println("progress: " + progress));
+        params.print_progress = CBool.FALSE;
+        //params.initial_prompt = "and so my fellow Americans um, like";
+
+        try {
+            audioInputStream.read(b);
+
+            for (int i = 0, j = 0; i < b.length; i += 2, j++) {
+                int intSample = (int) (b[i + 1]) << 8 | (int) (b[i]) & 0xFF;
+                floats[j] = intSample / 32767.0f;
+            }
+
+            List<WhisperSegment> segments = whisper.fullTranscribeWithTime(params, floats);
+            assertTrue(segments.size() > 0, "The size of segments should be greater than 0");
+            for (WhisperSegment segment : segments) {
+                System.out.println(segment);
+            }
+        } finally {
+            audioInputStream.close();
+        }
+    }
+
 }

bindings/javascript/package.json

@@ -1,6 +1,6 @@
 {
   "name": "whisper.cpp",
-  "version": "1.4.3",
+  "version": "1.5.2",
   "description": "Whisper speech recognition",
   "main": "whisper.js",
   "scripts": {

coreml/whisper-encoder-impl.h

@@ -123,7 +123,7 @@ API_AVAILABLE(macos(12.0), ios(15.0), watchos(8.0), tvos(15.0)) __attribute__((v
 
 /**
     Make a prediction using the convenience interface
-    @param logmel_data as 1 × 80 × 3000 3-dimensional array of floats:
+    @param logmel_data as 1 × n_mel × 3000 3-dimensional array of floats:
     @param error If an error occurs, upon return contains an NSError object that describes the problem. If you are not interested in possible errors, pass in NULL.
     @return the prediction as whisper_encoder_implOutput
 */

coreml/whisper-encoder.h

@@ -3,6 +3,8 @@
 // Code is derived from the work of Github user @wangchou
 // ref: https://github.com/wangchou/callCoreMLFromCpp
 
+#include <stdint.h>
+
 #if __cplusplus
 extern "C" {
 #endif
@@ -14,6 +16,8 @@ void whisper_coreml_free(struct whisper_coreml_context * ctx);
 
 void whisper_coreml_encode(
         const whisper_coreml_context * ctx,
+                             int64_t   n_ctx,
+                             int64_t   n_mel,
                                float * mel,
                                float * out);
 

coreml/whisper-encoder.mm

@@ -48,13 +48,15 @@ void whisper_coreml_free(struct whisper_coreml_context * ctx) {
 
 void whisper_coreml_encode(
         const whisper_coreml_context * ctx,
+                             int64_t   n_ctx,
+                             int64_t   n_mel,
                                float * mel,
                                float * out) {
     MLMultiArray * inMultiArray = [
         [MLMultiArray alloc] initWithDataPointer: mel
-                                           shape: @[@1, @80, @3000]
+                                           shape: @[@1, @(n_mel), @(n_ctx)]
                                         dataType: MLMultiArrayDataTypeFloat32
-                                         strides: @[@(240000), @(3000), @1]
+                                         strides: @[@(n_ctx*n_mel), @(n_ctx), @1]
                                      deallocator: nil
                                            error: nil
     ];

examples/CMakeLists.txt

@@ -23,6 +23,7 @@ add_library(${TARGET} STATIC
     common.cpp
     common-ggml.h
     common-ggml.cpp
+    grammar-parser.cpp
     )
 
 include(DefaultTargetOptions)
@@ -64,6 +65,7 @@ elseif(CMAKE_JS_VERSION)
 else()
     add_subdirectory(main)
     add_subdirectory(stream)
+    add_subdirectory(server)
     add_subdirectory(command)
     add_subdirectory(bench)
     add_subdirectory(quantize)
@@ -71,3 +73,5 @@ else()
     add_subdirectory(talk-llama)
     add_subdirectory(lsp)
 endif()
+
+add_subdirectory(wchess)

examples/bench/bench.cpp

@@ -81,7 +81,7 @@ int whisper_bench_full(const whisper_params & params) {
     }
     // heat encoder
     if (int ret = whisper_encode(ctx, 0, params.n_threads) != 0) {
-        fprintf(stderr, "error: failed to encode model: %d\n", ret);
+        fprintf(stderr, "error: failed to encode: %d\n", ret);
         return 4;
     }
 
@@ -90,13 +90,13 @@ int whisper_bench_full(const whisper_params & params) {
 
     // prompt heat
     if (int ret = whisper_decode(ctx, tokens, 256, 0, params.n_threads) != 0) {
-        fprintf(stderr, "error: failed to encode model: %d\n", ret);
+        fprintf(stderr, "error: failed to decode: %d\n", ret);
         return 4;
     }
 
     // text-generation heat
     if (int ret = whisper_decode(ctx, tokens, 1, 256, params.n_threads) != 0) {
-        fprintf(stderr, "error: failed to encode model: %d\n", ret);
+        fprintf(stderr, "error: failed to decode: %d\n", ret);
         return 4;
     }
 
@@ -104,20 +104,30 @@ int whisper_bench_full(const whisper_params & params) {
 
     // actual run
     if (int ret = whisper_encode(ctx, 0, params.n_threads) != 0) {
-        fprintf(stderr, "error: failed to encode model: %d\n", ret);
+        fprintf(stderr, "error: failed to encode: %d\n", ret);
         return 4;
     }
 
-    for (int i = 0; i < 16; i++) {
-        if (int ret = whisper_decode(ctx, tokens, 256, 0, params.n_threads) != 0) {
-            fprintf(stderr, "error: failed to encode model: %d\n", ret);
+    // text-generation
+    for (int i = 0; i < 256; i++) {
+        if (int ret = whisper_decode(ctx, tokens, 1, i, params.n_threads) != 0) {
+            fprintf(stderr, "error: failed to decode: %d\n", ret);
             return 4;
         }
     }
 
-    for (int i = 0; i < 256; i++) {
-        if (int ret = whisper_decode(ctx, tokens, 1, i, params.n_threads) != 0) {
-            fprintf(stderr, "error: failed to encode model: %d\n", ret);
+    // batched decoding
+    for (int i = 0; i < 64; i++) {
+        if (int ret = whisper_decode(ctx, tokens, 5, 0, params.n_threads) != 0) {
+            fprintf(stderr, "error: failed to decode: %d\n", ret);
+            return 4;
+        }
+    }
+
+    // prompt processing
+    for (int i = 0; i < 16; i++) {
+        if (int ret = whisper_decode(ctx, tokens, 256, 0, params.n_threads) != 0) {
+            fprintf(stderr, "error: failed to decode: %d\n", ret);
             return 4;
         }
     }

examples/command/command.cpp

@@ -9,6 +9,7 @@
 #include "common-sdl.h"
 #include "common.h"
 #include "whisper.h"
+#include "grammar-parser.h"
 
 #include <sstream>
 #include <cassert>
@@ -21,6 +22,11 @@
 #include <vector>
 #include <map>
 
+bool file_exists(const std::string & fname) {
+    std::ifstream f(fname.c_str());
+    return f.good();
+}
+
 // command-line parameters
 struct whisper_params {
     int32_t n_threads  = std::min(4, (int32_t) std::thread::hardware_concurrency());
@@ -30,8 +36,12 @@ struct whisper_params {
     int32_t max_tokens = 32;
     int32_t audio_ctx  = 0;
 
-    float vad_thold    = 0.6f;
-    float freq_thold   = 100.0f;
+    float vad_thold  = 0.6f;
+    float freq_thold = 100.0f;
+
+    float grammar_penalty = 100.0f;
+
+    grammar_parser::parse_state grammar_parsed;
 
     bool speed_up      = false;
     bool translate     = false;
@@ -45,6 +55,8 @@ struct whisper_params {
     std::string fname_out;
     std::string commands;
     std::string prompt;
+    std::string context;
+    std::string grammar;
 };
 
 void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@@ -75,6 +87,9 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         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 if (arg == "-ctx" || arg == "--context")       { params.context       = argv[++i]; }
+        else if (                 arg == "--grammar")       { params.grammar       = argv[++i]; }
+        else if (                 arg == "--grammar-penalty") { params.grammar_penalty = std::stof(argv[++i]); }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -109,16 +124,30 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     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, "  -ctx,       --context        [%-7s] sample text to help the transcription\n",       params.context.c_str());
+    fprintf(stderr, "  --grammar GRAMMAR            [%-7s] GBNF grammar to guide decoding\n",              params.grammar.c_str());
+    fprintf(stderr, "  --grammar-penalty N          [%-7.1f] scales down logits of nongrammar tokens\n",   params.grammar_penalty);
     fprintf(stderr, "\n");
 }
 
-std::string transcribe(whisper_context * ctx, const whisper_params & params, const std::vector<float> & pcmf32, float & prob, int64_t & t_ms) {
+std::string transcribe(
+                 whisper_context * ctx,
+            const whisper_params & params,
+        const std::vector<float> & pcmf32,
+               const std::string & grammar_rule,
+                           float & logprob_min,
+                           float & logprob_sum,
+                             int & n_tokens,
+                         int64_t & t_ms) {
     const auto t_start = std::chrono::high_resolution_clock::now();
 
-    prob = 0.0f;
+    logprob_min = 0.0f;
+    logprob_sum = 0.0f;
+    n_tokens    = 0;
     t_ms = 0;
 
-    whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
+    //whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
+    whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_BEAM_SEARCH);
 
     wparams.print_progress   = false;
     wparams.print_special    = params.print_special;
@@ -126,19 +155,41 @@ std::string transcribe(whisper_context * ctx, const whisper_params & params, con
     wparams.print_timestamps = !params.no_timestamps;
     wparams.translate        = params.translate;
     wparams.no_context       = true;
+    wparams.no_timestamps    = params.no_timestamps;
     wparams.single_segment   = true;
     wparams.max_tokens       = params.max_tokens;
     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.audio_ctx = params.audio_ctx;
+    wparams.speed_up  = params.speed_up;
+
+    wparams.temperature     = 0.4f;
+    wparams.temperature_inc = 1.0f;
+    wparams.greedy.best_of  = 5;
+
+    wparams.beam_search.beam_size = 5;
+
+    wparams.initial_prompt = params.context.data();
+
+    const auto & grammar_parsed = params.grammar_parsed;
+    auto grammar_rules = grammar_parsed.c_rules();
+
+    if (!params.grammar_parsed.rules.empty() && !grammar_rule.empty()) {
+        if (grammar_parsed.symbol_ids.find(grammar_rule) == grammar_parsed.symbol_ids.end()) {
+            fprintf(stderr, "%s: warning: grammar rule '%s' not found - skipping grammar sampling\n", __func__, grammar_rule.c_str());
+        } else {
+            wparams.grammar_rules   = grammar_rules.data();
+            wparams.n_grammar_rules = grammar_rules.size();
+            wparams.i_start_rule    = grammar_parsed.symbol_ids.at(grammar_rule);
+            wparams.grammar_penalty = params.grammar_penalty;
+        }
+    }
 
     if (whisper_full(ctx, wparams, pcmf32.data(), pcmf32.size()) != 0) {
         return "";
     }
 
-    int prob_n = 0;
     std::string result;
 
     const int n_segments = whisper_full_n_segments(ctx);
@@ -147,19 +198,17 @@ std::string transcribe(whisper_context * ctx, const whisper_params & params, con
 
         result += text;
 
-        const int n_tokens = whisper_full_n_tokens(ctx, i);
-        for (int j = 0; j < n_tokens; ++j) {
+        const int n = whisper_full_n_tokens(ctx, i);
+        for (int j = 0; j < n; ++j) {
             const auto token = whisper_full_get_token_data(ctx, i, j);
 
-            prob += token.p;
-            ++prob_n;
+            if(token.plog > 0.0f) exit(0);
+            logprob_min = std::min(logprob_min, token.plog);
+            logprob_sum += token.plog;
+            ++n_tokens;
         }
     }
 
-    if (prob_n > 0) {
-        prob /= prob_n;
-    }
-
     const auto t_end = std::chrono::high_resolution_clock::now();
     t_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t_end - t_start).count();
 
@@ -250,7 +299,7 @@ int process_command_list(struct whisper_context * ctx, audio_async &audio, const
         fprintf(stderr, " ]\n");
     }
 
-    std::string  k_prompt = "select one from the available words: ";
+    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 += ", ";
@@ -418,7 +467,9 @@ int always_prompt_transcription(struct whisper_context * ctx, audio_async & audi
     bool is_running = true;
     bool ask_prompt = true;
 
-    float prob = 0.0f;
+    float logprob_min = 0.0f;
+    float logprob_sum = 0.0f;
+    int   n_tokens    = 0;
 
     std::vector<float> pcmf32_cur;
 
@@ -456,7 +507,7 @@ int always_prompt_transcription(struct whisper_context * ctx, audio_async & audi
                 // detect the commands
                 audio.get(params.command_ms, pcmf32_cur);
 
-                const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob, t_ms));
+                const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, "", logprob_min, logprob_sum, n_tokens, t_ms));
 
                 const auto words = get_words(txt);
 
@@ -492,18 +543,27 @@ int always_prompt_transcription(struct whisper_context * ctx, audio_async & audi
 
 // general-purpose mode
 // freely transcribe the voice into text
-int process_general_transcription(struct whisper_context * ctx, audio_async &audio, const whisper_params &params) {
+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;
+    float logprob_min0 = 0.0f;
+    float logprob_min  = 0.0f;
+
+    float logprob_sum0 = 0.0f;
+    float logprob_sum  = 0.0f;
+
+    int n_tokens0 = 0;
+    int n_tokens  = 0;
 
     std::vector<float> pcmf32_cur;
     std::vector<float> pcmf32_prompt;
 
-    const std::string k_prompt = "Ok Whisper, start listening for commands.";
+    std::string k_prompt = "Ok Whisper, start listening for commands.";
+    if (!params.prompt.empty()) {
+        k_prompt = params.prompt;
+    }
 
     fprintf(stderr, "\n");
     fprintf(stderr, "%s: general-purpose mode\n", __func__);
@@ -536,9 +596,11 @@ int process_general_transcription(struct whisper_context * ctx, audio_async &aud
                     // wait for activation phrase
                     audio.get(params.prompt_ms, pcmf32_cur);
 
-                    const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob0, t_ms));
+                    const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, "prompt", logprob_min0, logprob_sum0, n_tokens0, 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 p = 100.0f * std::exp(logprob_min0);
+
+                    fprintf(stdout, "%s: Heard '%s%s%s', (t = %d ms, p = %.2f%%)\n", __func__, "\033[1m", txt.c_str(), "\033[0m", (int) t_ms, p);
 
                     const float sim = similarity(txt, k_prompt);
 
@@ -559,19 +621,30 @@ int process_general_transcription(struct whisper_context * ctx, audio_async &aud
                     // we have heard the activation phrase, now detect the commands
                     audio.get(params.command_ms, pcmf32_cur);
 
+                    //printf("len prompt:  %.4f\n", pcmf32_prompt.size() / (float) WHISPER_SAMPLE_RATE);
+                    //printf("len command: %.4f\n", pcmf32_cur.size() / (float) WHISPER_SAMPLE_RATE);
+
+                    // prepend 3 second of silence
+                    pcmf32_cur.insert(pcmf32_cur.begin(), 3.0f*WHISPER_SAMPLE_RATE, 0.0f);
+
                     // 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));
+                    const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, "root", logprob_min, logprob_sum, n_tokens, t_ms));
 
-                    prob = 100.0f*(prob - prob0);
+                    //const float p = 100.0f * std::exp((logprob - logprob0) / (n_tokens - n_tokens0));
+                    const float p = 100.0f * std::exp(logprob_min);
 
                     //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) {
+                    for (size_t n = 0.8*k_prompt.size(); n <= 1.2*k_prompt.size(); ++n) {
+                        if (n >= txt.size()) {
+                            break;
+                        }
+
                         const auto prompt = txt.substr(0, n);
 
                         const float sim = similarity(prompt, k_prompt);
@@ -584,9 +657,16 @@ int process_general_transcription(struct whisper_context * ctx, audio_async &aud
                         }
                     }
 
-                    const std::string command = ::trim(txt.substr(best_len));
+                    fprintf(stdout, "%s:   DEBUG: txt = '%s', prob = %.2f%%\n", __func__, txt.c_str(), p);
+                    if (best_len == 0) {
+                        fprintf(stdout, "%s: WARNING: command not recognized, try again\n", __func__);
+                    } else {
+                        // cut the prompt from the decoded text
+                        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, "%s: Command '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", command.c_str(), "\033[0m", (int) t_ms);
                     fprintf(stdout, "\n");
                 }
 
@@ -654,12 +734,36 @@ int main(int argc, char ** argv) {
 
     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 {
-        ret_val = process_general_transcription(ctx, audio, params);
+    if (!params.grammar.empty()) {
+        auto & grammar = params.grammar_parsed;
+        if (file_exists(params.grammar.c_str())) {
+            // read grammar from file
+            std::ifstream ifs(params.grammar.c_str());
+            const std::string txt = std::string((std::istreambuf_iterator<char>(ifs)), std::istreambuf_iterator<char>());
+            grammar = grammar_parser::parse(txt.c_str());
+        } else {
+            // read grammar from string
+            grammar = grammar_parser::parse(params.grammar.c_str());
+        }
+
+        // will be empty (default) if there are parse errors
+        if (grammar.rules.empty()) {
+            ret_val = 1;
+        } else {
+            fprintf(stderr, "%s: grammar:\n", __func__);
+            grammar_parser::print_grammar(stderr, grammar);
+            fprintf(stderr, "\n");
+        }
+    }
+
+    if (ret_val == 0) {
+        if (!params.commands.empty()) {
+            ret_val = process_command_list(ctx, audio, params);
+        } else if (!params.prompt.empty() && params.grammar_parsed.rules.empty()) {
+            ret_val = always_prompt_transcription(ctx, audio, params);
+        } else {
+            ret_val = process_general_transcription(ctx, audio, params);
+        }
     }
 
     audio.pause();

examples/common-ggml.cpp

@@ -9,6 +9,11 @@ static const std::map<std::string, enum ggml_ftype> GGML_FTYPE_MAP = {
     {"q5_0", GGML_FTYPE_MOSTLY_Q5_0},
     {"q5_1", GGML_FTYPE_MOSTLY_Q5_1},
     {"q8_0", GGML_FTYPE_MOSTLY_Q8_0},
+    {"q2_k", GGML_FTYPE_MOSTLY_Q2_K},
+    {"q3_k", GGML_FTYPE_MOSTLY_Q3_K},
+    {"q4_k", GGML_FTYPE_MOSTLY_Q4_K},
+    {"q5_k", GGML_FTYPE_MOSTLY_Q5_K},
+    {"q6_k", GGML_FTYPE_MOSTLY_Q6_K},
 };
 
 void ggml_print_ftypes(FILE * fp) {
@@ -48,15 +53,15 @@ bool ggml_common_quantize_0(
         case GGML_FTYPE_MOSTLY_Q5_0: qtype = GGML_TYPE_Q5_0; break;
         case GGML_FTYPE_MOSTLY_Q5_1: qtype = GGML_TYPE_Q5_1; break;
         case GGML_FTYPE_MOSTLY_Q8_0: qtype = GGML_TYPE_Q8_0; break;
+        case GGML_FTYPE_MOSTLY_Q2_K: qtype = GGML_TYPE_Q2_K; break;
+        case GGML_FTYPE_MOSTLY_Q3_K: qtype = GGML_TYPE_Q3_K; break;
+        case GGML_FTYPE_MOSTLY_Q4_K: qtype = GGML_TYPE_Q4_K; break;
+        case GGML_FTYPE_MOSTLY_Q5_K: qtype = GGML_TYPE_Q5_K; break;
+        case GGML_FTYPE_MOSTLY_Q6_K: qtype = GGML_TYPE_Q6_K; break;
         case GGML_FTYPE_UNKNOWN:
         case GGML_FTYPE_ALL_F32:
         case GGML_FTYPE_MOSTLY_F16:
         case GGML_FTYPE_MOSTLY_Q4_1_SOME_F16:
-        case GGML_FTYPE_MOSTLY_Q2_K:
-        case GGML_FTYPE_MOSTLY_Q3_K:
-        case GGML_FTYPE_MOSTLY_Q4_K:
-        case GGML_FTYPE_MOSTLY_Q5_K:
-        case GGML_FTYPE_MOSTLY_Q6_K:
                 {
                     fprintf(stderr, "%s: invalid model type %d\n", __func__, ftype);
                     return false;
@@ -167,24 +172,17 @@ bool ggml_common_quantize_0(
 
             switch ((ggml_type) ttype) {
                 case GGML_TYPE_Q4_0:
-                    {
-                        cur_size = ggml_quantize_q4_0(data_f32.data(), work.data(), nelements, ne[0], hist_cur.data());
-                    } break;
                 case GGML_TYPE_Q4_1:
-                    {
-                        cur_size = ggml_quantize_q4_1(data_f32.data(), work.data(), nelements, ne[0], hist_cur.data());
-                    } break;
                 case GGML_TYPE_Q5_0:
-                    {
-                        cur_size = ggml_quantize_q5_0(data_f32.data(), work.data(), nelements, ne[0], hist_cur.data());
-                    } break;
                 case GGML_TYPE_Q5_1:
-                    {
-                        cur_size = ggml_quantize_q5_1(data_f32.data(), work.data(), nelements, ne[0], hist_cur.data());
-                    } break;
                 case GGML_TYPE_Q8_0:
+                case GGML_TYPE_Q2_K:
+                case GGML_TYPE_Q3_K:
+                case GGML_TYPE_Q4_K:
+                case GGML_TYPE_Q5_K:
+                case GGML_TYPE_Q6_K:
                     {
-                        cur_size = ggml_quantize_q8_0(data_f32.data(), work.data(), nelements, ne[0], hist_cur.data());
+                        cur_size = ggml_quantize_chunk((ggml_type) ttype, data_f32.data(), work.data(), 0, nelements, hist_cur.data());
                     } break;
                 case GGML_TYPE_F32:
                 case GGML_TYPE_F16:
@@ -192,11 +190,6 @@ bool ggml_common_quantize_0(
                 case GGML_TYPE_I16:
                 case GGML_TYPE_I32:
                 case GGML_TYPE_Q8_1:
-                case GGML_TYPE_Q2_K:
-                case GGML_TYPE_Q3_K:
-                case GGML_TYPE_Q4_K:
-                case GGML_TYPE_Q5_K:
-                case GGML_TYPE_Q6_K:
                 case GGML_TYPE_Q8_K:
                 case GGML_TYPE_COUNT:
                     {

examples/common-sdl.cpp

@@ -139,10 +139,13 @@ void audio_async::callback(uint8_t * stream, int len) {
         return;
     }
 
-    const size_t n_samples = len / sizeof(float);
+    size_t n_samples = len / sizeof(float);
 
-    m_audio_new.resize(n_samples);
-    memcpy(m_audio_new.data(), stream, n_samples * sizeof(float));
+    if (n_samples > m_audio.size()) {
+        n_samples = m_audio.size();
+
+        stream += (len - (n_samples * sizeof(float)));
+    }
 
     //fprintf(stderr, "%s: %zu samples, pos %zu, len %zu\n", __func__, n_samples, m_audio_pos, m_audio_len);
 
@@ -153,7 +156,7 @@ void audio_async::callback(uint8_t * stream, int len) {
             const size_t n0 = m_audio.size() - m_audio_pos;
 
             memcpy(&m_audio[m_audio_pos], stream, n0 * sizeof(float));
-            memcpy(&m_audio[0], &stream[n0], (n_samples - n0) * sizeof(float));
+            memcpy(&m_audio[0], stream + n0 * sizeof(float), (n_samples - n0) * sizeof(float));
 
             m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
             m_audio_len = m_audio.size();

examples/common-sdl.h

@@ -41,7 +41,6 @@ private:
     std::mutex       m_mutex;
 
     std::vector<float> m_audio;
-    std::vector<float> m_audio_new;
     size_t             m_audio_pos = 0;
     size_t             m_audio_len = 0;
 };

examples/common.h

@@ -181,7 +181,7 @@ private:
     // It is assumed that PCM data is normalized to a range from -1 to 1
     bool write_audio(const float * data, size_t length) {
         for (size_t i = 0; i < length; ++i) {
-            const auto intSample = static_cast<const int16_t>(data[i] * 32767);
+            const int16_t intSample = data[i] * 32767;
             file.write(reinterpret_cast<const char *>(&intSample), sizeof(int16_t));
             dataSize += sizeof(int16_t);
         }

examples/grammar-parser.cpp

@@ -0,0 +1,423 @@
+#include "grammar-parser.h"
+#include <cstdint>
+#include <cwchar>
+#include <string>
+#include <utility>
+#include <stdexcept>
+#include <exception>
+
+namespace grammar_parser {
+    // NOTE: assumes valid utf8 (but checks for overrun)
+    // copied from whisper.cpp
+    std::pair<uint32_t, const char *> decode_utf8(const char * src) {
+        static const int lookup[] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 4 };
+        uint8_t  first_byte = static_cast<uint8_t>(*src);
+        uint8_t  highbits   = first_byte >> 4;
+        int      len        = lookup[highbits];
+        uint8_t  mask       = (1 << (8 - len)) - 1;
+        uint32_t value      = first_byte & mask;
+        const char * end    = src + len; // may overrun!
+        const char * pos    = src + 1;
+        for ( ; pos < end && *pos; pos++) {
+            value = (value << 6) + (static_cast<uint8_t>(*pos) & 0x3F);
+        }
+        return std::make_pair(value, pos);
+    }
+
+    uint32_t get_symbol_id(parse_state & state, const char * src, size_t len) {
+        uint32_t next_id = static_cast<uint32_t>(state.symbol_ids.size());
+        auto result = state.symbol_ids.insert(std::make_pair(std::string(src, len), next_id));
+        return result.first->second;
+    }
+
+    uint32_t generate_symbol_id(parse_state & state, const std::string & base_name) {
+        uint32_t next_id = static_cast<uint32_t>(state.symbol_ids.size());
+        state.symbol_ids[base_name + '_' + std::to_string(next_id)] = next_id;
+        return next_id;
+    }
+
+    void add_rule(
+            parse_state & state,
+            uint32_t      rule_id,
+            const std::vector<whisper_grammar_element> & rule) {
+        if (state.rules.size() <= rule_id) {
+            state.rules.resize(rule_id + 1);
+        }
+        state.rules[rule_id] = rule;
+    }
+
+    bool is_word_char(char c) {
+        return ('a' <= c && c <= 'z') || ('A' <= c && c <= 'Z') || c == '-' || ('0' <= c && c <= '9');
+    }
+
+    std::pair<uint32_t, const char *> parse_hex(const char * src, int size) {
+        const char * pos   = src;
+        const char * end   = src + size;
+        uint32_t     value = 0;
+        for ( ; pos < end && *pos; pos++) {
+            value <<= 4;
+            char c = *pos;
+            if ('a' <= c && c <= 'f') {
+                value += c - 'a' + 10;
+            } else if ('A' <= c && c <= 'F') {
+                value += c - 'A' + 10;
+            } else if ('0' <= c && c <= '9') {
+                value += c - '0';
+            } else {
+                break;
+            }
+        }
+        if (pos != end) {
+            throw std::runtime_error("expecting " + std::to_string(size) + " hex chars at " + src);
+        }
+        return std::make_pair(value, pos);
+    }
+
+    const char * parse_space(const char * src, bool newline_ok) {
+        const char * pos = src;
+        while (*pos == ' ' || *pos == '\t' || *pos == '#' ||
+                (newline_ok && (*pos == '\r' || *pos == '\n'))) {
+            if (*pos == '#') {
+                while (*pos && *pos != '\r' && *pos != '\n') {
+                    pos++;
+                }
+            } else {
+                pos++;
+            }
+        }
+        return pos;
+    }
+
+    const char * parse_name(const char * src) {
+        const char * pos = src;
+        while (is_word_char(*pos)) {
+            pos++;
+        }
+        if (pos == src) {
+            throw std::runtime_error(std::string("expecting name at ") + src);
+        }
+        return pos;
+    }
+
+    std::pair<uint32_t, const char *> parse_char(const char * src) {
+        if (*src == '\\') {
+            switch (src[1]) {
+                case 'x': return parse_hex(src + 2, 2);
+                case 'u': return parse_hex(src + 2, 4);
+                case 'U': return parse_hex(src + 2, 8);
+                case 't': return std::make_pair('\t', src + 2);
+                case 'r': return std::make_pair('\r', src + 2);
+                case 'n': return std::make_pair('\n', src + 2);
+                case '\\':
+                case '"':
+                case '[':
+                case ']':
+                    return std::make_pair(src[1], src + 2);
+                default:
+                    throw std::runtime_error(std::string("unknown escape at ") + src);
+            }
+        } else if (*src) {
+            return decode_utf8(src);
+        }
+        throw std::runtime_error("unexpected end of input");
+    }
+
+    const char * parse_alternates(
+            parse_state       & state,
+            const char        * src,
+            const std::string & rule_name,
+            uint32_t            rule_id,
+            bool                is_nested);
+
+    const char * parse_sequence(
+            parse_state                        & state,
+            const char                         * src,
+            const std::string                  & rule_name,
+            std::vector<whisper_grammar_element> & out_elements,
+            bool                                 is_nested) {
+        size_t last_sym_start = out_elements.size();
+        const char * pos = src;
+        while (*pos) {
+            if (*pos == '"') { // literal string
+                pos++;
+                last_sym_start = out_elements.size();
+                while (*pos != '"') {
+                    auto char_pair = parse_char(pos);
+                         pos       = char_pair.second;
+                    out_elements.push_back({WHISPER_GRETYPE_CHAR, char_pair.first});
+                }
+                pos = parse_space(pos + 1, is_nested);
+            } else if (*pos == '[') { // char range(s)
+                pos++;
+                enum whisper_gretype start_type = WHISPER_GRETYPE_CHAR;
+                if (*pos == '^') {
+                    pos++;
+                    start_type = WHISPER_GRETYPE_CHAR_NOT;
+                }
+                last_sym_start = out_elements.size();
+                while (*pos != ']') {
+                    auto char_pair = parse_char(pos);
+                         pos       = char_pair.second;
+                    enum whisper_gretype type = last_sym_start < out_elements.size()
+                        ? WHISPER_GRETYPE_CHAR_ALT
+                        : start_type;
+
+                    out_elements.push_back({type, char_pair.first});
+                    if (pos[0] == '-' && pos[1] != ']') {
+                        auto endchar_pair = parse_char(pos + 1);
+                             pos          = endchar_pair.second;
+                        out_elements.push_back({WHISPER_GRETYPE_CHAR_RNG_UPPER, endchar_pair.first});
+                    }
+                }
+                pos = parse_space(pos + 1, is_nested);
+            } else if (is_word_char(*pos)) { // rule reference
+                const char * name_end    = parse_name(pos);
+                uint32_t     ref_rule_id = get_symbol_id(state, pos, name_end - pos);
+                pos = parse_space(name_end, is_nested);
+                last_sym_start = out_elements.size();
+                out_elements.push_back({WHISPER_GRETYPE_RULE_REF, ref_rule_id});
+            } else if (*pos == '(') { // grouping
+                // parse nested alternates into synthesized rule
+                pos = parse_space(pos + 1, true);
+                uint32_t sub_rule_id = generate_symbol_id(state, rule_name);
+                pos = parse_alternates(state, pos, rule_name, sub_rule_id, true);
+                last_sym_start = out_elements.size();
+                // output reference to synthesized rule
+                out_elements.push_back({WHISPER_GRETYPE_RULE_REF, sub_rule_id});
+                if (*pos != ')') {
+                    throw std::runtime_error(std::string("expecting ')' at ") + pos);
+                }
+                pos = parse_space(pos + 1, is_nested);
+            } else if (*pos == '*' || *pos == '+' || *pos == '?') { // repetition operator
+                if (last_sym_start == out_elements.size()) {
+                    throw std::runtime_error(std::string("expecting preceeding item to */+/? at ") + pos);
+                }
+
+                // apply transformation to previous symbol (last_sym_start to end) according to
+                // rewrite rules:
+                // S* --> S' ::= S S' |
+                // S+ --> S' ::= S S' | S
+                // S? --> S' ::= S |
+                uint32_t sub_rule_id = generate_symbol_id(state, rule_name);
+                std::vector<whisper_grammar_element> sub_rule;
+                // add preceding symbol to generated rule
+                sub_rule.insert(
+                    sub_rule.end(), out_elements.begin() + last_sym_start, out_elements.end());
+                if (*pos == '*' || *pos == '+') {
+                    // cause generated rule to recurse
+                    sub_rule.push_back({WHISPER_GRETYPE_RULE_REF, sub_rule_id});
+                }
+                // mark start of alternate def
+                sub_rule.push_back({WHISPER_GRETYPE_ALT, 0});
+                if (*pos == '+') {
+                    // add preceding symbol as alternate only for '+' (otherwise empty)
+                    sub_rule.insert(
+                        sub_rule.end(), out_elements.begin() + last_sym_start, out_elements.end());
+                }
+                sub_rule.push_back({WHISPER_GRETYPE_END, 0});
+                add_rule(state, sub_rule_id, sub_rule);
+
+                // in original rule, replace previous symbol with reference to generated rule
+                out_elements.resize(last_sym_start);
+                out_elements.push_back({WHISPER_GRETYPE_RULE_REF, sub_rule_id});
+
+                pos = parse_space(pos + 1, is_nested);
+            } else {
+                break;
+            }
+        }
+        return pos;
+    }
+
+    const char * parse_alternates(
+            parse_state       & state,
+            const char        * src,
+            const std::string & rule_name,
+            uint32_t            rule_id,
+            bool                is_nested) {
+        std::vector<whisper_grammar_element> rule;
+        const char * pos = parse_sequence(state, src, rule_name, rule, is_nested);
+        while (*pos == '|') {
+            rule.push_back({WHISPER_GRETYPE_ALT, 0});
+            pos = parse_space(pos + 1, true);
+            pos = parse_sequence(state, pos, rule_name, rule, is_nested);
+        }
+        rule.push_back({WHISPER_GRETYPE_END, 0});
+        add_rule(state, rule_id, rule);
+        return pos;
+    }
+
+    const char * parse_rule(parse_state & state, const char * src) {
+        const char * name_end = parse_name(src);
+        const char * pos      = parse_space(name_end, false);
+        size_t       name_len = name_end - src;
+        uint32_t     rule_id  = get_symbol_id(state, src, name_len);
+        const std::string name(src, name_len);
+
+        if (!(pos[0] == ':' && pos[1] == ':' && pos[2] == '=')) {
+            throw std::runtime_error(std::string("expecting ::= at ") + pos);
+        }
+        pos = parse_space(pos + 3, true);
+
+        pos = parse_alternates(state, pos, name, rule_id, false);
+
+        if (*pos == '\r') {
+            pos += pos[1] == '\n' ? 2 : 1;
+        } else if (*pos == '\n') {
+            pos++;
+        } else if (*pos) {
+            throw std::runtime_error(std::string("expecting newline or end at ") + pos);
+        }
+        return parse_space(pos, true);
+    }
+
+    parse_state parse(const char * src) {
+        try {
+            parse_state state;
+            const char * pos = parse_space(src, true);
+            while (*pos) {
+                pos = parse_rule(state, pos);
+            }
+            return state;
+        } catch (const std::exception & err) {
+            fprintf(stderr, "%s: error parsing grammar: %s\n", __func__, err.what());
+            return parse_state();
+        }
+    }
+
+    void print_grammar_char(FILE * file, uint32_t c) {
+        if (0x20 <= c && c <= 0x7f) {
+            fprintf(file, "%c", static_cast<char>(c));
+        } else {
+            // cop out of encoding UTF-8
+            fprintf(file, "<U+%04X>", c);
+        }
+    }
+
+    bool is_char_element(whisper_grammar_element elem) {
+        switch (elem.type) {
+            case WHISPER_GRETYPE_CHAR:           return true;
+            case WHISPER_GRETYPE_CHAR_NOT:       return true;
+            case WHISPER_GRETYPE_CHAR_ALT:       return true;
+            case WHISPER_GRETYPE_CHAR_RNG_UPPER: return true;
+            default:                           return false;
+        }
+    }
+
+    void print_rule_binary(FILE * file, const std::vector<whisper_grammar_element> & rule) {
+        for (auto elem : rule) {
+            switch (elem.type) {
+                case WHISPER_GRETYPE_END:            fprintf(file, "END");            break;
+                case WHISPER_GRETYPE_ALT:            fprintf(file, "ALT");            break;
+                case WHISPER_GRETYPE_RULE_REF:       fprintf(file, "RULE_REF");       break;
+                case WHISPER_GRETYPE_CHAR:           fprintf(file, "CHAR");           break;
+                case WHISPER_GRETYPE_CHAR_NOT:       fprintf(file, "CHAR_NOT");       break;
+                case WHISPER_GRETYPE_CHAR_RNG_UPPER: fprintf(file, "CHAR_RNG_UPPER"); break;
+                case WHISPER_GRETYPE_CHAR_ALT:       fprintf(file, "CHAR_ALT");       break;
+            }
+            switch (elem.type) {
+                case WHISPER_GRETYPE_END:
+                case WHISPER_GRETYPE_ALT:
+                case WHISPER_GRETYPE_RULE_REF:
+                    fprintf(file, "(%u) ", elem.value);
+                    break;
+                case WHISPER_GRETYPE_CHAR:
+                case WHISPER_GRETYPE_CHAR_NOT:
+                case WHISPER_GRETYPE_CHAR_RNG_UPPER:
+                case WHISPER_GRETYPE_CHAR_ALT:
+                    fprintf(file, "(\"");
+                    print_grammar_char(file, elem.value);
+                    fprintf(file, "\") ");
+                    break;
+            }
+        }
+        fprintf(file, "\n");
+    }
+
+    void print_rule(
+            FILE     * file,
+            uint32_t   rule_id,
+            const std::vector<whisper_grammar_element> & rule,
+            const std::map<uint32_t, std::string>    & symbol_id_names) {
+        if (rule.empty() || rule.back().type != WHISPER_GRETYPE_END) {
+            throw std::runtime_error(
+                "malformed rule, does not end with WHISPER_GRETYPE_END: " + std::to_string(rule_id));
+        }
+        fprintf(file, "%s ::= ", symbol_id_names.at(rule_id).c_str());
+        for (size_t i = 0, end = rule.size() - 1; i < end; i++) {
+            whisper_grammar_element elem = rule[i];
+            switch (elem.type) {
+                case WHISPER_GRETYPE_END:
+                    throw std::runtime_error(
+                        "unexpected end of rule: " + std::to_string(rule_id) + "," +
+                        std::to_string(i));
+                case WHISPER_GRETYPE_ALT:
+                    fprintf(file, "| ");
+                    break;
+                case WHISPER_GRETYPE_RULE_REF:
+                    fprintf(file, "%s ", symbol_id_names.at(elem.value).c_str());
+                    break;
+                case WHISPER_GRETYPE_CHAR:
+                    fprintf(file, "[");
+                    print_grammar_char(file, elem.value);
+                    break;
+                case WHISPER_GRETYPE_CHAR_NOT:
+                    fprintf(file, "[^");
+                    print_grammar_char(file, elem.value);
+                    break;
+                case WHISPER_GRETYPE_CHAR_RNG_UPPER:
+                    if (i == 0 || !is_char_element(rule[i - 1])) {
+                        throw std::runtime_error(
+                            "WHISPER_GRETYPE_CHAR_RNG_UPPER without preceding char: " +
+                            std::to_string(rule_id) + "," + std::to_string(i));
+                    }
+                    fprintf(file, "-");
+                    print_grammar_char(file, elem.value);
+                    break;
+                case WHISPER_GRETYPE_CHAR_ALT:
+                    if (i == 0 || !is_char_element(rule[i - 1])) {
+                        throw std::runtime_error(
+                            "WHISPER_GRETYPE_CHAR_ALT without preceding char: " +
+                            std::to_string(rule_id) + "," + std::to_string(i));
+                    }
+                    print_grammar_char(file, elem.value);
+                    break;
+            }
+            if (is_char_element(elem)) {
+                switch (rule[i + 1].type) {
+                    case WHISPER_GRETYPE_CHAR_ALT:
+                    case WHISPER_GRETYPE_CHAR_RNG_UPPER:
+                        break;
+                    default:
+                        fprintf(file, "] ");
+                }
+            }
+        }
+        fprintf(file, "\n");
+    }
+
+    void print_grammar(FILE * file, const parse_state & state) {
+        try {
+            std::map<uint32_t, std::string> symbol_id_names;
+            for (auto kv : state.symbol_ids) {
+                symbol_id_names[kv.second] = kv.first;
+            }
+            for (size_t i = 0, end = state.rules.size(); i < end; i++) {
+                // fprintf(file, "%zu: ", i);
+                // print_rule_binary(file, state.rules[i]);
+                print_rule(file, uint32_t(i), state.rules[i], symbol_id_names);
+                // fprintf(file, "\n");
+            }
+        } catch (const std::exception & err) {
+            fprintf(stderr, "\n%s: error printing grammar: %s\n", __func__, err.what());
+        }
+    }
+
+    std::vector<const whisper_grammar_element *> parse_state::c_rules() const{
+        std::vector<const whisper_grammar_element *> ret;
+        for (const auto & rule : rules) {
+            ret.push_back(rule.data());
+        }
+        return ret;
+    }
+}

examples/grammar-parser.h

@@ -0,0 +1,29 @@
+// Implements a parser for an extended Backus-Naur form (BNF), producing the
+// binary context-free grammar format specified by whisper.h. Supports character
+// ranges, grouping, and repetition operators. As an example, a grammar for
+// arithmetic might look like:
+//
+// root  ::= expr
+// expr  ::= term ([-+*/] term)*
+// term  ::= num | "(" space expr ")" space
+// num   ::= [0-9]+ space
+// space ::= [ \t\n]*
+
+#pragma once
+#include "whisper.h"
+#include <vector>
+#include <map>
+#include <cstdint>
+#include <string>
+
+namespace grammar_parser {
+    struct parse_state {
+        std::map<std::string, uint32_t>                   symbol_ids;
+        std::vector<std::vector<whisper_grammar_element>> rules;
+
+        std::vector<const whisper_grammar_element *>      c_rules() const;
+    };
+
+    parse_state parse(const char * src);
+    void print_grammar(FILE * file, const parse_state & state);
+}

examples/helpers.js

@@ -22,6 +22,7 @@ var printTextarea = (function() {
 async function clearCache() {
     if (confirm('Are you sure you want to clear the cache?\nAll the models will be downloaded again.')) {
         indexedDB.deleteDatabase(dbName);
+        location.reload();
     }
 }
 

examples/livestream.sh

@@ -48,7 +48,7 @@ if [ -n "$3" ]; then
 fi
 
 # Whisper models
-models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large-v2" "large" )
+models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large-v1" "large-v2" "large-v3" )
 
 # list available models
 function list_models {

examples/main/README.md

@@ -17,28 +17,37 @@ options:
   -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
+  -sow,      --split-on-word     [false  ] split on word rather than on token
   -bo N,     --best-of N         [5      ] number of best candidates to keep
-  -bs N,     --beam-size N       [-1     ] beam size for beam search
+  -bs N,     --beam-size N       [5      ] 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)
+  -debug,    --debug-mode        [false  ] enable debug mode (eg. dump log_mel)
   -tr,       --translate         [false  ] translate from source language to english
   -di,       --diarize           [false  ] stereo audio diarization
+  -tdrz,     --tinydiarize       [false  ] enable tinydiarize (requires a tdrz model)
   -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
+  -olrc,     --output-lrc        [false  ] output result in a lrc file
   -owts,     --output-words      [false  ] output script for generating karaoke video
+  -fp,       --font-path         [/System/Library/Fonts/Supplemental/Courier New Bold.ttf] path to a monospace font for karaoke video
   -ocsv,     --output-csv        [false  ] output result in a CSV file
   -oj,       --output-json       [false  ] output result in a JSON file
+  -ojf,      --output-json-full  [false  ] include more information in the JSON 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
+  -nt,       --no-timestamps     [false  ] do not print timestamps
   -l LANG,   --language LANG     [en     ] spoken language ('auto' for auto-detect)
+  -dl,       --detect-language   [false  ] exit after automatically detecting language
              --prompt PROMPT     [       ] initial prompt
   -m FNAME,  --model FNAME       [models/ggml-base.en.bin] model path
   -f FNAME,  --file FNAME        [       ] input WAV file path
+  -oved D,   --ov-e-device DNAME [CPU    ] the OpenVINO device used for encode inference
+  -ls,       --log-score         [false  ] log best decoder scores of tokens
+  -ng,       --no-gpu            [false  ] disable GPU
 ```

examples/main/main.cpp

@@ -62,8 +62,8 @@ struct whisper_params {
     int32_t progress_step =  5;
     int32_t max_context  = -1;
     int32_t max_len      =  0;
-    int32_t best_of      =  2;
-    int32_t beam_size    = -1;
+    int32_t best_of      = whisper_full_default_params(WHISPER_SAMPLING_GREEDY).greedy.best_of;
+    int32_t beam_size    = whisper_full_default_params(WHISPER_SAMPLING_BEAM_SEARCH).beam_search.beam_size;
 
     float word_thold    =  0.01f;
     float entropy_thold =  2.40f;
@@ -165,8 +165,8 @@ 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_inp.emplace_back(argv[++i]); }
         else if (arg == "-oved" || arg == "--ov-e-device")     { params.openvino_encode_device = argv[++i]; }
-        else if (arg == "-ls"   || arg == "--log-score")       { params.log_score = true; }
-        else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu = false; }
+        else if (arg == "-ls"   || arg == "--log-score")       { params.log_score       = true; }
+        else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu         = false; }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -925,9 +925,9 @@ int main(int argc, char ** argv) {
             if (params.detect_language) {
                 params.language = "auto";
             }
-            fprintf(stderr, "%s: processing '%s' (%d samples, %.1f sec), %d threads, %d processors, lang = %s, task = %s, %stimestamps = %d ...\n",
+            fprintf(stderr, "%s: processing '%s' (%d samples, %.1f sec), %d threads, %d processors, %d beams + best of %d, lang = %s, task = %s, %stimestamps = %d ...\n",
                     __func__, fname_inp.c_str(), int(pcmf32.size()), float(pcmf32.size())/WHISPER_SAMPLE_RATE,
-                    params.n_threads, params.n_processors,
+                    params.n_threads, params.n_processors, params.beam_size, params.best_of,
                     params.language.c_str(),
                     params.translate ? "translate" : "transcribe",
                     params.tinydiarize ? "tdrz = 1, " : "",

examples/server/CMakeLists.txt

@@ -0,0 +1,12 @@
+set(TARGET server)
+add_executable(${TARGET} server.cpp httplib.h json.hpp)
+
+include(DefaultTargetOptions)
+
+target_link_libraries(${TARGET} PRIVATE common whisper ${CMAKE_THREAD_LIBS_INIT})
+
+# Check if the compiler is MinGW
+if(MINGW)
+    # Link the necessary libraries for SSL and Winsock
+    target_link_libraries(${TARGET} PRIVATE -lcrypt32 -lssl -lcrypto -lws2_32)
+endif()

examples/server/README.md

@@ -0,0 +1,68 @@
+# whisper.cpp http server
+
+Simple http server. WAV Files are passed to the inference model via http requests.
+
+https://github.com/ggerganov/whisper.cpp/assets/1991296/e983ee53-8741-4eb5-9048-afe5e4594b8f
+
+## Usage
+
+```
+./server -h
+
+usage: ./bin/server [options]
+
+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
+  -sow,      --split-on-word     [false  ] split on word rather than on token
+  -bo N,     --best-of N         [2      ] 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
+  -debug,    --debug-mode        [false  ] enable debug mode (eg. dump log_mel)
+  -tr,       --translate         [false  ] translate from source language to english
+  -di,       --diarize           [false  ] stereo audio diarization
+  -tdrz,     --tinydiarize       [false  ] enable tinydiarize (requires a tdrz model)
+  -nf,       --no-fallback       [false  ] do not use temperature fallback while decoding
+  -ps,       --print-special     [false  ] print special tokens
+  -pc,       --print-colors      [false  ] print colors
+  -pr,       --print-realtime    [false  ] print output in realtime
+  -pp,       --print-progress    [false  ] print progress
+  -nt,       --no-timestamps     [false  ] do not print timestamps
+  -l LANG,   --language LANG     [en     ] spoken language ('auto' for auto-detect)
+  -dl,       --detect-language   [false  ] exit after automatically detecting language
+             --prompt PROMPT     [       ] initial prompt
+  -m FNAME,  --model FNAME       [models/ggml-base.en.bin] model path
+  -oved D,   --ov-e-device DNAME [CPU    ] the OpenVINO device used for encode inference
+  --host HOST,                   [127.0.0.1] Hostname/ip-adress for the server
+  --port PORT,                   [8080   ] Port number for the server
+  --convert,                     [false  ] Convert audio to WAV, requires ffmpeg on the server
+```
+
+> [!WARNING]  
+> **Do not run the server example with administrative privileges and ensure it's operated in a sandbox environment, especially since it involves risky operations like accepting user file uploads and using ffmpeg for format conversions. Always validate and sanitize inputs to guard against potential security threats.**
+
+## request examples
+
+**/inference**
+```
+curl 127.0.0.1:8080/inference \
+-H "Content-Type: multipart/form-data" \
+-F file="@<file-path>" \
+-F temperature="0.2" \
+-F response-format="json"
+```
+
+**/load**
+```
+curl 127.0.0.1:8080/load \
+-H "Content-Type: multipart/form-data" \
+-F model="<path-to-model-file>"
+```

examples/server/server.cpp

@@ -0,0 +1,811 @@
+#include "common.h"
+
+#include "whisper.h"
+#include "httplib.h"
+#include "json.hpp"
+
+#include <cmath>
+#include <fstream>
+#include <cstdio>
+#include <string>
+#include <thread>
+#include <vector>
+#include <cstring>
+#include <sstream>
+
+#if defined(_MSC_VER)
+#pragma warning(disable: 4244 4267) // possible loss of data
+#endif
+
+using namespace httplib;
+using json = nlohmann::json;
+
+namespace {
+
+// Terminal color map. 10 colors grouped in ranges [0.0, 0.1, ..., 0.9]
+// Lowest is red, middle is yellow, highest is green.
+const std::vector<std::string> k_colors = {
+    "\033[38;5;196m", "\033[38;5;202m", "\033[38;5;208m", "\033[38;5;214m", "\033[38;5;220m",
+    "\033[38;5;226m", "\033[38;5;190m", "\033[38;5;154m", "\033[38;5;118m", "\033[38;5;82m",
+};
+
+// output formats
+const std::string json_format   = "json";
+const std::string text_format   = "text";
+const std::string srt_format    = "srt";
+const std::string vjson_format  = "verbose_json";
+const std::string vtt_format    = "vtt";
+
+struct server_params
+{
+    std::string hostname = "127.0.0.1";
+    std::string public_path = "examples/server/public";
+
+    int32_t port          = 8080;
+    int32_t read_timeout  = 600;
+    int32_t write_timeout = 600;
+    
+    bool ffmpeg_converter = false;
+};
+
+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 progress_step =  5;
+    int32_t max_context  = -1;
+    int32_t max_len      =  0;
+    int32_t best_of      =  2;
+    int32_t beam_size    = -1;
+
+    float word_thold    =  0.01f;
+    float entropy_thold =  2.40f;
+    float logprob_thold = -1.00f;
+    float userdef_temp  =  0.20f;
+
+    bool speed_up        = false;
+    bool debug_mode      = false;
+    bool translate       = false;
+    bool detect_language = false;
+    bool diarize         = false;
+    bool tinydiarize     = false;
+    bool split_on_word   = false;
+    bool no_fallback     = false;
+    bool print_special   = false;
+    bool print_colors    = false;
+    bool print_realtime  = false;
+    bool print_progress  = false;
+    bool no_timestamps   = false;
+    bool use_gpu         = true;
+
+    std::string language        = "en";
+    std::string prompt          = "";
+    std::string font_path       = "/System/Library/Fonts/Supplemental/Courier New Bold.ttf";
+    std::string model           = "models/ggml-base.en.bin";
+
+    std::string response_format     = json_format;
+
+    // [TDRZ] speaker turn string
+    std::string tdrz_speaker_turn = " [SPEAKER_TURN]"; // TODO: set from command line
+
+    std::string openvino_encode_device = "CPU";
+};
+
+//  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)));
+}
+
+bool is_file_exist(const char *fileName)
+{
+    std::ifstream infile(fileName);
+    return infile.good();
+}
+
+void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params,
+                         const server_params& sparams) {
+    fprintf(stderr, "\n");
+    fprintf(stderr, "usage: %s [options] \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, "  -sow,      --split-on-word     [%-7s] split on word rather than on token\n",             params.split_on_word ? "true" : "false");
+    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, "  -debug,    --debug-mode        [%-7s] enable debug mode (eg. dump log_mel)\n",           params.debug_mode ? "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, "  -tdrz,     --tinydiarize       [%-7s] enable tinydiarize (requires a tdrz model)\n",     params.tinydiarize ? "true" : "false");
+    fprintf(stderr, "  -nf,       --no-fallback       [%-7s] do not use temperature fallback while decoding\n", params.no_fallback ? "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, "  -pr,       --print-realtime    [%-7s] print output in realtime\n",                       params.print_realtime ? "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 ? "true" : "false");
+    fprintf(stderr, "  -l LANG,   --language LANG     [%-7s] spoken language ('auto' for auto-detect)\n",       params.language.c_str());
+    fprintf(stderr, "  -dl,       --detect-language   [%-7s] exit after automatically detecting language\n",    params.detect_language ? "true" : "false");
+    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, "  -oved D,   --ov-e-device DNAME [%-7s] the OpenVINO device used for encode inference\n",  params.openvino_encode_device.c_str());
+    // server params
+    fprintf(stderr, "  --host HOST,                   [%-7s] Hostname/ip-adress for the server\n", sparams.hostname.c_str());
+    fprintf(stderr, "  --port PORT,                   [%-7d] Port number for the server\n", sparams.port);
+    fprintf(stderr, "  --public PATH,                 [%-7s] Path to the public folder\n", sparams.public_path.c_str());
+    fprintf(stderr, "  --convert,                     [%-7s] Convert audio to WAV, requires ffmpeg on the server", sparams.ffmpeg_converter ? "true" : "false");
+    fprintf(stderr, "\n");
+}
+
+bool whisper_params_parse(int argc, char ** argv, whisper_params & params, server_params & sparams) {
+    for (int i = 1; i < argc; i++) {
+        std::string arg = argv[i];
+
+        if (arg == "-h" || arg == "--help") {
+            whisper_print_usage(argc, argv, params, sparams);
+            exit(0);
+        }
+        else if (arg == "-t"    || arg == "--threads")         { params.n_threads       = std::stoi(argv[++i]); }
+        else if (arg == "-p"    || arg == "--processors")      { params.n_processors    = std::stoi(argv[++i]); }
+        else if (arg == "-ot"   || arg == "--offset-t")        { params.offset_t_ms     = std::stoi(argv[++i]); }
+        else if (arg == "-on"   || arg == "--offset-n")        { params.offset_n        = std::stoi(argv[++i]); }
+        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 == "-debug"|| arg == "--debug-mode")      { params.debug_mode      = true; }
+        else if (arg == "-tr"   || arg == "--translate")       { params.translate       = true; }
+        else if (arg == "-di"   || arg == "--diarize")         { params.diarize         = true; }
+        else if (arg == "-tdrz" || arg == "--tinydiarize")     { params.tinydiarize     = true; }
+        else if (arg == "-sow"  || arg == "--split-on-word")   { params.split_on_word   = true; }
+        else if (arg == "-nf"   || arg == "--no-fallback")     { params.no_fallback     = true; }
+        else if (arg == "-fp"   || arg == "--font-path")       { params.font_path       = 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 == "-pr"   || arg == "--print-realtime")  { params.print_realtime  = true; }
+        else if (arg == "-pp"   || arg == "--print-progress")  { params.print_progress  = true; }
+        else if (arg == "-nt"   || arg == "--no-timestamps")   { params.no_timestamps   = true; }
+        else if (arg == "-l"    || arg == "--language")        { params.language        = argv[++i]; }
+        else if (arg == "-dl"   || arg == "--detect-language") { params.detect_language = true; }
+        else if (                  arg == "--prompt")          { params.prompt          = argv[++i]; }
+        else if (arg == "-m"    || arg == "--model")           { params.model           = argv[++i]; }
+        else if (arg == "-oved" || arg == "--ov-e-device")     { params.openvino_encode_device = argv[++i]; }
+        else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu         = false; }
+        // server params
+        else if (                  arg == "--port")            { sparams.port        = std::stoi(argv[++i]); }
+        else if (                  arg == "--host")            { sparams.hostname    = argv[++i]; }
+        else if (                  arg == "--public")          { sparams.public_path = argv[++i]; }
+        else if (                  arg == "--convert")         { sparams.ffmpeg_converter     = true; }
+        else {
+            fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
+            whisper_print_usage(argc, argv, params, sparams);
+            exit(0);
+        }
+    }
+
+    return true;
+}
+
+struct whisper_print_user_data {
+    const whisper_params * params;
+
+    const std::vector<std::vector<float>> * pcmf32s;
+    int progress_prev;
+};
+
+void check_ffmpeg_availibility() {
+    int result = system("ffmpeg -version");
+
+    if (result == 0) {
+        std::cout << "ffmpeg is available." << std::endl;
+    } else {
+        // ffmpeg is not available
+        std::cout << "ffmpeg is not found. Please ensure that ffmpeg is installed ";
+        std::cout << "and that its executable is included in your system's PATH. ";
+        exit(0);
+    }
+}
+
+bool convert_to_wav(const std::string & temp_filename, std::string & error_resp) {
+    std::ostringstream cmd_stream;
+    std::string converted_filename_temp = temp_filename + "_temp.wav";
+    cmd_stream << "ffmpeg -i \"" << temp_filename << "\" -ar 16000 -ac 1 -c:a pcm_s16le \"" << converted_filename_temp << "\" 2>&1";
+    std::string cmd = cmd_stream.str();
+
+    int status = std::system(cmd.c_str());
+    if (status != 0) {
+        error_resp = "{\"error\":\"FFmpeg conversion failed.\"}";
+        return false;
+    }
+
+    // Remove the original file
+    if (remove(temp_filename.c_str()) != 0) {
+        error_resp = "{\"error\":\"Failed to remove the original file.\"}";
+        return false;
+    }
+
+    // Rename the temporary file to match the original filename
+    if (rename(converted_filename_temp.c_str(), temp_filename.c_str()) != 0) {
+        error_resp = "{\"error\":\"Failed to rename the temporary file.\"}";
+        return false;
+    }
+    return true;
+}
+
+std::string estimate_diarization_speaker(std::vector<std::vector<float>> pcmf32s, int64_t t0, int64_t t1, bool id_only = false) {
+    std::string speaker = "";
+    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 = "0";
+    } else if (energy1 > 1.1*energy0) {
+        speaker = "1";
+    } else {
+        speaker = "?";
+    }
+
+    //printf("is0 = %lld, is1 = %lld, energy0 = %f, energy1 = %f, speaker = %s\n", is0, is1, energy0, energy1, speaker.c_str());
+
+    if (!id_only) {
+        speaker.insert(0, "(speaker ");
+        speaker.append(")");
+    }
+
+    return speaker;
+}
+
+void whisper_print_progress_callback(struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, int progress, void * user_data) {
+    int progress_step = ((whisper_print_user_data *) user_data)->params->progress_step;
+    int * progress_prev  = &(((whisper_print_user_data *) user_data)->progress_prev);
+    if (progress >= *progress_prev + progress_step) {
+        *progress_prev += progress_step;
+        fprintf(stderr, "%s: progress = %3d%%\n", __func__, progress);
+    }
+}
+
+void whisper_print_segment_callback(struct whisper_context * ctx, struct whisper_state * /*state*/, 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 = 0;
+    int64_t t1 = 0;
+
+    // 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) {
+            speaker = estimate_diarization_speaker(pcmf32s, t0, t1);
+        }
+
+        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);
+        }
+
+        if (params.tinydiarize) {
+            if (whisper_full_get_segment_speaker_turn_next(ctx, i)) {
+                printf("%s", params.tdrz_speaker_turn.c_str());
+            }
+        }
+
+        // with timestamps or speakers: each segment on new line
+        if (!params.no_timestamps || params.diarize) {
+            printf("\n");
+        }
+        fflush(stdout);
+    }
+}
+
+std::string output_str(struct whisper_context * ctx, const whisper_params & params, std::vector<std::vector<float>> pcmf32s) {
+    std::stringstream result;
+    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);
+        std::string speaker = "";
+
+        if (params.diarize && pcmf32s.size() == 2)
+        {
+            const int64_t t0 = whisper_full_get_segment_t0(ctx, i);
+            const int64_t t1 = whisper_full_get_segment_t1(ctx, i);
+            speaker = estimate_diarization_speaker(pcmf32s, t0, t1);
+        }
+
+        result << speaker << text << "\n";
+    }
+    return result.str();
+}
+
+void get_req_parameters(const Request & req, whisper_params & params)
+{
+    // user model configu.has_fileion
+    if (req.has_file("offset-t"))
+    {
+        params.offset_t_ms = std::stoi(req.get_file_value("offset-t").content);
+    }
+    if (req.has_file("offset-n"))
+    {
+        params.offset_n = std::stoi(req.get_file_value("offset-n").content);
+    }
+    if (req.has_file("duration"))
+    {
+        params.duration_ms = std::stoi(req.get_file_value("duration").content);
+    }
+    if (req.has_file("max-context"))
+    {
+        params.max_context = std::stoi(req.get_file_value("max-context").content);
+    }
+    if (req.has_file("prompt"))
+    {
+        params.prompt = req.get_file_value("prompt").content;
+    }
+    if (req.has_file("response-format"))
+    {
+        params.response_format = req.get_file_value("response-format").content;
+    }
+    if (req.has_file("temperature"))
+    {
+        params.userdef_temp = std::stof(req.get_file_value("temperature").content);
+    }
+}
+
+}  // namespace
+
+int main(int argc, char ** argv) {
+    whisper_params params;
+    server_params sparams;
+
+    std::mutex whisper_mutex;
+
+    if (whisper_params_parse(argc, argv, params, sparams) == false) {
+        whisper_print_usage(argc, argv, params, sparams);
+        return 1;
+    }
+
+    if (params.language != "auto" && whisper_lang_id(params.language.c_str()) == -1) {
+        fprintf(stderr, "error: unknown language '%s'\n", params.language.c_str());
+        whisper_print_usage(argc, argv, params, sparams);
+        exit(0);
+    }
+
+    if (params.diarize && params.tinydiarize) {
+        fprintf(stderr, "error: cannot use both --diarize and --tinydiarize\n");
+        whisper_print_usage(argc, argv, params, sparams);
+        exit(0);
+    }
+
+    if (sparams.ffmpeg_converter) {
+        check_ffmpeg_availibility();
+    }
+    // whisper init
+    struct whisper_context_params cparams;
+    cparams.use_gpu = params.use_gpu;
+
+    struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+
+    if (ctx == nullptr) {
+        fprintf(stderr, "error: failed to initialize whisper context\n");
+        return 3;
+    }
+
+    // initialize openvino encoder. this has no effect on whisper.cpp builds that don't have OpenVINO configured
+    whisper_ctx_init_openvino_encoder(ctx, nullptr, params.openvino_encode_device.c_str(), nullptr);
+
+    Server svr;
+    svr.set_default_headers({{"Server", "whisper.cpp"},
+                             {"Access-Control-Allow-Origin", "*"},
+                             {"Access-Control-Allow-Headers", "content-type"}});
+
+    std::string const default_content = "<html>hello</html>";
+
+    // this is only called if no index.html is found in the public --path
+    svr.Get("/", [&default_content](const Request &, Response &res){
+        res.set_content(default_content, "text/html");
+        return false;
+    });
+
+    svr.Post("/inference", [&](const Request &req, Response &res){
+        // acquire whisper model mutex lock
+        whisper_mutex.lock();
+
+        // first check user requested fields of the request
+        if (!req.has_file("file"))
+        {
+            fprintf(stderr, "error: no 'file' field in the request\n");
+            const std::string error_resp = "{\"error\":\"no 'file' field in the request\"}";
+            res.set_content(error_resp, "application/json");
+            whisper_mutex.unlock();
+            return;
+        }
+        auto audio_file = req.get_file_value("file");
+
+        // check non-required fields
+        get_req_parameters(req, params);
+
+        std::string filename{audio_file.filename};
+        printf("Received request: %s\n", filename.c_str());
+
+        // audio arrays
+        std::vector<float> pcmf32;               // mono-channel F32 PCM
+        std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
+
+        // write to temporary file
+        const std::string temp_filename = "whisper_server_temp_file.wav";
+        std::ofstream temp_file{temp_filename, std::ios::binary};
+        temp_file << audio_file.content;
+        temp_file.close();
+
+        // if file is not wav, convert to wav
+        
+        if (sparams.ffmpeg_converter) {
+            std::string error_resp = "{\"error\":\"Failed to execute ffmpeg command.\"}";
+            const bool is_converted = convert_to_wav(temp_filename, error_resp);
+            if (!is_converted) {
+                res.set_content(error_resp, "application/json");
+                whisper_mutex.unlock();
+                return;
+            }
+        }
+
+        // read wav content into pcmf32
+        if (!::read_wav(temp_filename, pcmf32, pcmf32s, params.diarize)) {
+            fprintf(stderr, "error: failed to read WAV file '%s'\n", temp_filename.c_str());
+            const std::string error_resp = "{\"error\":\"failed to read WAV file\"}";
+            res.set_content(error_resp, "application/json");
+            std::remove(temp_filename.c_str());
+            whisper_mutex.unlock();
+            return;
+        }
+        // remove temp file
+        std::remove(temp_filename.c_str());
+
+        printf("Successfully loaded %s\n", filename.c_str());
+
+        // 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__);
+                }
+            }
+            if (params.detect_language) {
+                params.language = "auto";
+            }
+            fprintf(stderr, "%s: processing '%s' (%d samples, %.1f sec), %d threads, %d processors, lang = %s, task = %s, %stimestamps = %d ...\n",
+                    __func__, filename.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.tinydiarize ? "tdrz = 1, " : "",
+                    params.no_timestamps ? 0 : 1);
+
+            fprintf(stderr, "\n");
+        }
+
+        // run the inference
+        {
+            printf("Running whisper.cpp inference on %s\n", filename.c_str());
+            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.detect_language  = params.detect_language;
+            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.thold_pt         = params.word_thold;
+            wparams.max_len          = params.max_len == 0 ? 60 : params.max_len;
+            wparams.split_on_word    = params.split_on_word;
+
+            wparams.speed_up         = params.speed_up;
+            wparams.debug_mode       = params.debug_mode;
+
+            wparams.tdrz_enable      = params.tinydiarize; // [TDRZ]
+
+            wparams.initial_prompt   = params.prompt.c_str();
+
+            wparams.greedy.best_of        = params.best_of;
+            wparams.beam_search.beam_size = params.beam_size;
+
+            wparams.temperature_inc  = params.userdef_temp;
+            wparams.entropy_thold    = params.entropy_thold;
+            wparams.logprob_thold    = params.logprob_thold;
+
+            whisper_print_user_data user_data = { &params, &pcmf32s, 0 };
+
+            // this callback is called on each new segment
+            if (params.print_realtime) {
+                wparams.new_segment_callback           = whisper_print_segment_callback;
+                wparams.new_segment_callback_user_data = &user_data;
+            }
+
+            if (wparams.print_progress) {
+                wparams.progress_callback           = whisper_print_progress_callback;
+                wparams.progress_callback_user_data = &user_data;
+            }
+
+            // examples for abort mechanism
+            // in examples below, 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*/, struct whisper_state * /*state*/, void * user_data) {
+                    bool is_aborted = *(bool*)user_data;
+                    return !is_aborted;
+                };
+                wparams.encoder_begin_callback_user_data = &is_aborted;
+            }
+
+            // the callback is called before every computation - if it returns true, the computation is aborted
+            {
+                static bool is_aborted = false; // NOTE: this should be atomic to avoid data race
+
+                wparams.abort_callback = [](void * user_data) {
+                    bool is_aborted = *(bool*)user_data;
+                    return is_aborted;
+                };
+                wparams.abort_callback_user_data = &is_aborted;
+            }
+
+            if (whisper_full_parallel(ctx, wparams, pcmf32.data(), pcmf32.size(), params.n_processors) != 0) {
+                fprintf(stderr, "%s: failed to process audio\n", argv[0]);
+                const std::string error_resp = "{\"error\":\"failed to process audio\"}";
+                res.set_content(error_resp, "application/json");
+                whisper_mutex.unlock();
+                return;
+            }
+        }
+
+        // return results to user
+        if (params.response_format == text_format)
+        {
+            std::string results = output_str(ctx, params, pcmf32s);
+            res.set_content(results.c_str(), "text/html");
+        }
+        else if (params.response_format == srt_format)
+        {
+            std::stringstream ss;
+            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);
+                std::string speaker = "";
+
+                if (params.diarize && pcmf32s.size() == 2)
+                {
+                    speaker = estimate_diarization_speaker(pcmf32s, t0, t1);
+                }
+
+                ss << i + 1 + params.offset_n << "\n";
+                ss << to_timestamp(t0, true) << " --> " << to_timestamp(t1, true) << "\n";
+                ss << speaker << text << "\n\n";
+            }
+            res.set_content(ss.str(), "application/x-subrip");
+        } else if (params.response_format == vtt_format) {
+            std::stringstream ss;
+
+            ss << "WEBVTT\n\n";
+
+            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);
+                std::string speaker = "";
+
+                if (params.diarize && pcmf32s.size() == 2)
+                {
+                    speaker = estimate_diarization_speaker(pcmf32s, t0, t1, true);
+                    speaker.insert(0, "<v Speaker");
+                    speaker.append(">");
+                }
+
+                ss << to_timestamp(t0) << " --> " << to_timestamp(t1) << "\n";
+                ss << speaker << text << "\n\n";
+            }
+            res.set_content(ss.str(), "text/vtt");
+        }
+        // TODO add more output formats
+        else
+        {
+            std::string results = output_str(ctx, params, pcmf32s);
+            json jres = json{
+                {"text", results}
+            };
+            res.set_content(jres.dump(-1, ' ', false, json::error_handler_t::replace),
+                            "application/json");
+        }
+
+        // return whisper model mutex lock
+        whisper_mutex.unlock();
+    });
+    svr.Post("/load", [&](const Request &req, Response &res){
+        whisper_mutex.lock();
+        if (!req.has_file("model"))
+        {
+            fprintf(stderr, "error: no 'model' field in the request\n");
+            const std::string error_resp = "{\"error\":\"no 'model' field in the request\"}";
+            res.set_content(error_resp, "application/json");
+            whisper_mutex.unlock();
+            return;
+        }
+        std::string model = req.get_file_value("model").content;
+        if (!is_file_exist(model.c_str()))
+        {
+            fprintf(stderr, "error: 'model': %s not found!\n", model.c_str());
+            const std::string error_resp = "{\"error\":\"model not found!\"}";
+            res.set_content(error_resp, "application/json");
+            whisper_mutex.unlock();
+            return;
+        }
+
+        // clean up
+        whisper_free(ctx);
+
+        // whisper init
+        ctx = whisper_init_from_file_with_params(model.c_str(), cparams);
+
+        // TODO perhaps load prior model here instead of exit
+        if (ctx == nullptr) {
+            fprintf(stderr, "error: model init  failed, no model loaded must exit\n");
+            exit(1);
+        }
+
+        // initialize openvino encoder. this has no effect on whisper.cpp builds that don't have OpenVINO configured
+        whisper_ctx_init_openvino_encoder(ctx, nullptr, params.openvino_encode_device.c_str(), nullptr);
+
+        const std::string success = "Load was successful!";
+        res.set_content(success, "application/text");
+
+        // check if the model is in the file system
+        whisper_mutex.unlock();
+    });
+
+    svr.set_exception_handler([](const Request &, Response &res, std::exception_ptr ep) {
+        const char fmt[] = "500 Internal Server Error\n%s";
+        char buf[BUFSIZ];
+        try {
+            std::rethrow_exception(std::move(ep));
+        } catch (std::exception &e) {
+            snprintf(buf, sizeof(buf), fmt, e.what());
+        } catch (...) {
+            snprintf(buf, sizeof(buf), fmt, "Unknown Exception");
+        }
+        res.set_content(buf, "text/plain");
+        res.status = 500;
+    });
+
+    svr.set_error_handler([](const Request &, Response &res) {
+        if (res.status == 400) {
+            res.set_content("Invalid request", "text/plain");
+        } else if (res.status != 500) {
+            res.set_content("File Not Found", "text/plain");
+            res.status = 404;
+        }
+    });
+
+    // set timeouts and change hostname and port
+    svr.set_read_timeout(sparams.read_timeout);
+    svr.set_write_timeout(sparams.write_timeout);
+
+    if (!svr.bind_to_port(sparams.hostname, sparams.port))
+    {
+        fprintf(stderr, "\ncouldn't bind to server socket: hostname=%s port=%d\n\n",
+                sparams.hostname.c_str(), sparams.port);
+        return 1;
+    }
+
+    // Set the base directory for serving static files
+    svr.set_base_dir(sparams.public_path);
+
+    // to make it ctrl+clickable:
+    printf("\nwhisper server listening at http://%s:%d\n\n", sparams.hostname.c_str(), sparams.port);
+
+    if (!svr.listen_after_bind())
+    {
+        return 1;
+    }
+
+    whisper_print_timings(ctx);
+    whisper_free(ctx);
+
+    return 0;
+}

examples/talk-llama/CMakeLists.txt

@@ -18,6 +18,11 @@ if (WHISPER_SDL2)
         ../../ggml-quants.c
         ../../whisper.cpp)
 
+    if(WIN32)
+    # It requires Windows 8.1 or later for PrefetchVirtualMemory
+    target_compile_definitions(${TARGET} PRIVATE -D_WIN32_WINNT=0x0602)
+    endif()
+
     target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS} ../../)
     target_link_libraries(${TARGET} PRIVATE ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
 

examples/talk-llama/talk-llama.cpp

@@ -53,6 +53,7 @@ struct whisper_params {
     int32_t capture_id = -1;
     int32_t max_tokens = 32;
     int32_t audio_ctx  = 0;
+    int32_t n_gpu_layers = 999;
 
     float vad_thold  = 0.6f;
     float freq_thold = 100.0f;
@@ -90,6 +91,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
         else if (arg == "-c"   || arg == "--capture")        { params.capture_id     = std::stoi(argv[++i]); }
         else if (arg == "-mt"  || arg == "--max-tokens")     { params.max_tokens     = std::stoi(argv[++i]); }
         else if (arg == "-ac"  || arg == "--audio-ctx")      { params.audio_ctx      = std::stoi(argv[++i]); }
+        else if (arg == "-ngl" || arg == "--n-gpu-layers")   { params.n_gpu_layers   = std::stoi(argv[++i]); }
         else if (arg == "-vth" || arg == "--vad-thold")      { params.vad_thold      = std::stof(argv[++i]); }
         else if (arg == "-fth" || arg == "--freq-thold")     { params.freq_thold     = std::stof(argv[++i]); }
         else if (arg == "-su"  || arg == "--speed-up")       { params.speed_up       = true; }
@@ -134,6 +136,7 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -c ID,    --capture ID     [%-7d] capture device ID\n",                           params.capture_id);
     fprintf(stderr, "  -mt N,    --max-tokens N   [%-7d] maximum number of tokens per audio chunk\n",    params.max_tokens);
     fprintf(stderr, "  -ac N,    --audio-ctx N    [%-7d] audio context size (0 - all)\n",                params.audio_ctx);
+    fprintf(stderr, "  -ngl N,   --n-gpu-layers N [%-7d] number of layers to store in VRAM\n",           params.n_gpu_layers);
     fprintf(stderr, "  -vth N,   --vad-thold N    [%-7.2f] voice activity detection threshold\n",        params.vad_thold);
     fprintf(stderr, "  -fth N,   --freq-thold N   [%-7.2f] high-pass frequency cutoff\n",                params.freq_thold);
     fprintf(stderr, "  -su,      --speed-up       [%-7s] speed up audio by x2 (reduced accuracy)\n",     params.speed_up ? "true" : "false");
@@ -248,7 +251,7 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
-    if (whisper_lang_id(params.language.c_str()) == -1) {
+    if (params.language != "auto" && whisper_lang_id(params.language.c_str()) == -1) {
         fprintf(stderr, "error: unknown language '%s'\n", params.language.c_str());
         whisper_print_usage(argc, argv, params);
         exit(0);
@@ -268,6 +271,8 @@ int main(int argc, char ** argv) {
     auto lmparams = llama_model_default_params();
     if (!params.use_gpu) {
         lmparams.n_gpu_layers = 0;
+    } else {
+        lmparams.n_gpu_layers = params.n_gpu_layers;
     }
 
     struct llama_model * model_llama = llama_load_model_from_file(params.model_llama.c_str(), lmparams);
@@ -681,8 +686,8 @@ int main(int argc, char ** argv) {
                     }
                 }
 
-                text_to_speak = ::replace(text_to_speak, "\"", "");
-                int ret = system((params.speak + " " + std::to_string(voice_id) + " \"" + text_to_speak + "\"").c_str());
+                text_to_speak = ::replace(text_to_speak, "'", "'\"'\"'");
+                int ret = system((params.speak + " " + std::to_string(voice_id) + " '" + text_to_speak + "'").c_str());
                 if (ret != 0) {
                     fprintf(stderr, "%s: failed to speak\n", __func__);
                 }

examples/talk/gpt-2.cpp

@@ -121,13 +121,13 @@ bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab &
             return false;
         }
 
-        std::string word;
+        char word[129];
+
         for (int i = 0; i < n_vocab; i++) {
             uint32_t len;
             fin.read((char *) &len, sizeof(len));
-
-            word.resize(len);
-            fin.read((char *) word.data(), len);
+            word[len] = '\0';
+            fin.read((char *) word, len);
 
             vocab.token_to_id[word] = i;
             vocab.id_to_token[i] = word;

examples/twitch.sh

@@ -21,7 +21,7 @@ help()
     echo "Usage: ./twitch.sh -s [step] -m [model] -t [threads] [url]"
     echo "options:"
     echo "-s       Step in seconds (default is $step)."
-    echo "-m       Choose model, options are: 'tiny.en' 'tiny' 'base.en' 'base' 'small.en' 'small' 'medium.en' 'medium' 'large-v1' 'large-v2' 'large' (default is '$model')."
+    echo "-m       Choose model, options are: 'tiny.en' 'tiny' 'base.en' 'base' 'small.en' 'small' 'medium.en' 'medium' 'large-v1' 'large-v2' 'large-v3' (default is '$model')."
     echo "-t       Number of threads to use."
     echo "-h       Print this help page."
     echo

examples/wchess/CMakeLists.txt

@@ -0,0 +1,9 @@
+set(CMAKE_CXX_STANDARD 11)
+
+add_subdirectory(libwchess)
+
+if (EMSCRIPTEN)
+    add_subdirectory(wchess.wasm)
+else()
+    add_subdirectory(wchess.cmd)
+endif()

examples/wchess/README.md

@@ -0,0 +1,40 @@
+# wchess
+
+Voice-controlled chess using Whisper
+
+Online demo: https://whisper.ggerganov.com/wchess/
+
+https://github.com/ggerganov/whisper.cpp/assets/1991296/c2b2f03c-9684-49f3-8106-357d2d4e67fa
+
+## Command-line tool
+
+```bash
+mkdir build && cd build
+cmake -DWHISPER_SDL2=1 ..
+make -j
+
+./bin/wchess -m ../models/ggml-base.en.bin
+
+Move: start
+
+a b c d e f g h
+r n b q k b n r 8
+p p p p p p p p 7
+. * . * . * . * 6
+* . * . * . * . 5
+. * . * . * . * 4
+* . * . * . * . 3
+P P P P P P P P 2
+R N B Q K B N R 1
+
+White's turn
+[(l)isten/(p)ause/(q)uit]: 
+```
+
+## TODO
+
+- Improve web-browser audio capture - sometimes it does not record the voice properly
+- Add support for more languages by making the generated grammar string multi-lingual
+- Fix bugs in the chess moves logic
+
+PRs welcome!

examples/wchess/libwchess/CMakeLists.txt

@@ -0,0 +1,19 @@
+add_library(wchess-core STATIC
+    WChess.cpp
+    WChess.h
+    Chessboard.cpp
+    Chessboard.h
+)
+
+target_link_libraries(wchess-core
+    PUBLIC
+    whisper
+    common
+)
+
+target_include_directories(wchess-core
+    PUBLIC
+    "$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>"
+)
+
+# add_executable(test-chessboard test-chessboard.cpp Chessboard.cpp)

examples/wchess/libwchess/Chessboard.cpp

@@ -0,0 +1,803 @@
+#include "Chessboard.h"
+
+#include <array>
+#include <vector>
+#include <algorithm>
+#include <cstring>
+#include <set>
+#include <list>
+#include <chrono>
+
+namespace {
+constexpr std::array<const char*, 64> positions = {
+    "a1", "b1", "c1", "d1", "e1", "f1", "g1", "h1",
+    "a2", "b2", "c2", "d2", "e2", "f2", "g2", "h2",
+    "a3", "b3", "c3", "d3", "e3", "f3", "g3", "h3",
+    "a4", "b4", "c4", "d4", "e4", "f4", "g4", "h4",
+    "a5", "b5", "c5", "d5", "e5", "f5", "g5", "h5",
+    "a6", "b6", "c6", "d6", "e6", "f6", "g6", "h6",
+    "a7", "b7", "c7", "d7", "e7", "f7", "g7", "h7",
+    "a8", "b8", "c8", "d8", "e8", "f8", "g8", "h8",
+};
+constexpr char INVALID_POS = positions.size();
+constexpr int R = 0; // rank index
+constexpr int F = 1; // file index
+#define FILE (c[F] - '1')
+#define RANK (c[R] - 'a')
+constexpr char operator ""_P(const char * c, size_t size) {
+    return size < 2 || RANK < 0 || RANK > 7 ||
+        FILE < 0 || FILE > 7 ? INVALID_POS : FILE * 8 + RANK;
+}
+#undef FILE
+#undef RANK
+
+struct sview {
+    const char * ptr = nullptr;
+    size_t size = 0;
+
+    sview() = default;
+    sview(const char * p, size_t s) : ptr(p), size(s) {}
+    sview(const std::string& s) : ptr(s.data()), size(s.size()) {}
+
+    size_t find(char del, size_t pos) {
+        while (pos < size && ptr[pos] != del) ++pos;
+        return pos < size ? pos : std::string::npos;
+    }
+};
+
+std::vector<sview> split(sview str, char del) {
+    std::vector<sview> res;
+    size_t cur = 0;
+    size_t last = 0;
+    while (cur != std::string::npos) {
+        if (str.ptr[last] == ' ') {
+            ++last;
+            continue;
+        }
+        cur = str.find(del, last);
+        size_t len = cur == std::string::npos ? str.size - last : cur - last;
+        res.emplace_back(str.ptr + last, len);
+        last = cur + 1;
+    }
+    return res;
+}
+
+char strToPos(sview str) {
+    return operator ""_P(str.ptr, str.size);
+}
+
+constexpr std::array<const char*, 6> pieceNames =  {
+    "pawn", "knight", "bishop", "rook", "queen", "king",
+};
+
+static constexpr std::array<char, 6> blackShort =  {
+    'p', 'n', 'b', 'r', 'q', 'k',
+};
+static constexpr std::array<char, 6> whiteShort =  {
+    'P', 'N', 'B', 'R', 'Q', 'K',
+};
+
+char strToType(sview str) {
+    auto it = std::find_if(pieceNames.begin(), pieceNames.end(), [str] (const char* name) { return strncmp(name, str.ptr, str.size) == 0; });
+    return it != pieceNames.end() ? it - pieceNames.begin() : pieceNames.size();
+}
+
+// directions
+using Direction = std::array<char, 2>;
+
+constexpr Direction N   = {(char)  0, (char)  1};
+constexpr Direction NNE = {(char)  1, (char)  2};
+constexpr Direction NE  = {(char)  1, (char)  1};
+constexpr Direction ENE = {(char)  2, (char)  1};
+constexpr Direction E   = {(char)  1, (char)  0};
+constexpr Direction ESE = {(char)  2, (char) -1};
+constexpr Direction SE  = {(char)  1, (char) -1};
+constexpr Direction SSE = {(char)  1, (char) -2};
+constexpr Direction S   = {(char)  0, (char) -1};
+constexpr Direction SSW = {(char) -1, (char) -2};
+constexpr Direction SW  = {(char) -1, (char) -1};
+constexpr Direction WSW = {(char) -2, (char) -1};
+constexpr Direction W   = {(char) -1, (char)  0};
+constexpr Direction WNW = {(char) -2, (char)  1};
+constexpr Direction NW  = {(char) -1, (char)  1};
+constexpr Direction NNW = {(char) -1, (char)  2};
+
+char makeStep(char pos, const Direction& d) {
+    char next[2] = { char(positions[pos][R] + d[R]) , char(positions[pos][F] + d[F]) };
+    return strToPos(sview{next, sizeof(next)});
+}
+
+template<class Modifier>
+char traverse(char pos, const Direction& d, const Modifier& m, int count = 8) {
+    while (--count >= 0) {
+        pos = makeStep(pos, d);
+        if (pos == INVALID_POS || m(pos)) break;
+    }
+    return pos;
+}
+
+Direction normalize(const Direction& distance) {
+    //return {char((distance[R] > 0) - (distance[R] < 0)), char((distance[F] > 0) - (distance[F] < 0))};
+    const int drp = distance[R] > 0 ? 1 : 0;
+    const int drn = distance[R] < 0 ? 1 : 0;
+    const int dfp = distance[F] > 0 ? 1 : 0;
+    const int dfn = distance[F] < 0 ? 1 : 0;
+    return {char(drp - drn), char(dfp - dfn)};
+}
+
+struct Pin {
+    Direction d;
+    Piece* pinner;
+    Piece* pinned;
+};
+using Pins = std::list<Pin>;
+using Board = std::array<Piece*, 64>;
+
+std::vector<Direction> filter(const Direction& pin, std::initializer_list<Direction> directions) {
+    if (pin[R] == 0 && pin[F] == 0) return directions;
+    std::vector<Direction> result;
+    for (auto& d : directions) {
+        if ((d[R] == pin[R] || d[R] == -pin[R]) && (d[F] == pin[F] || d[F] == -pin[F])) result.push_back(d);
+    }
+    return result;
+}
+}
+
+class Piece {
+public:
+    enum Types : char {
+        Pawn,
+        Knight,
+        Bishop,
+        Rook,
+        Queen,
+        King,
+        //
+        NUM_PIECES
+    };
+
+    enum Colors : char {
+        White,
+        Black,
+    };
+
+    const char* name() const;
+    char initial() const;
+    Types type() const { return m_type; }
+    Colors color() const { return m_color; }
+    char pos() const { return m_pos; }
+    void setPos(char pos) {
+        m_pos = pos;
+        invalidate();
+    }
+    const char* coord() const;
+    const std::set<char>& allowed() const { return m_allowed; }
+    bool canReach(char pos) const;
+    virtual bool movePattern(char pos) const = 0;
+    void take();
+    virtual void reinit(const State& state) = 0;
+    void invalidate();
+protected:
+    Piece(Types type, Colors color, char pos, std::set<char> allowed)
+        : m_type(type), m_color(color), m_pos(pos), m_allowed(std::move(allowed)) {}
+    Piece(const Piece&) = delete;
+    ~Piece() = default;
+
+    const Types m_type;
+    const Colors m_color;
+    char m_pos;
+    std::set<char> m_allowed;
+    bool m_update = false;
+};
+
+struct Pawn : public Piece {
+    Pawn(Colors color, char pos, std::set<char> next) : Piece(Types::Pawn, color, pos, std::move(next)) {}
+
+    bool is_first_move() const {
+        return m_color ? coord()[F] == '7' : coord()[F] == '2';
+    }
+
+    virtual bool movePattern(char pos) const override {
+        if (m_pos == INVALID_POS) return false;
+        auto cur = coord();
+        auto next = positions[pos];
+        Direction distance = {char(next[R] - cur[R]), char(next[F] - cur[F])};
+        char forward = m_color ? -1 : 1;
+        return (forward == distance[F] && distance[R] * distance[R] <= 1)
+            || (is_first_move() && 2 * forward == distance[F] && distance[R] == 0);
+    }
+
+    virtual void reinit(const State& state) override;
+};
+
+struct Knight : public Piece {
+    Knight(Colors color, char pos, std::set<char> next) : Piece(Types::Knight, color, pos, std::move(next)) {}
+
+    virtual bool movePattern(char pos) const override {
+        if (m_pos == INVALID_POS) return false;
+        auto cur = coord();
+        auto next = positions[pos];
+        Direction diff = {char(next[R] - cur[R]), char(next[F] - cur[F])};
+        return diff[R]*diff[R] + diff[F]*diff[F] == 5;
+    }
+
+    virtual void reinit(const State& state) override;
+};
+
+struct Bishop : public Piece {
+    Bishop(Colors color, char pos) : Piece(Types::Bishop, color, pos, {}) {}
+
+    virtual bool movePattern(char pos) const override {
+        if (m_pos == INVALID_POS) return false;
+        auto cur = coord();
+        auto next = positions[pos];
+        return cur[R] - cur[F] == next[R] - next[F] || cur[R] + cur[F] == next[R] + next[F];
+    }
+
+    virtual void reinit(const State& state) override;
+};
+
+struct Rook : public Piece {
+    Rook(Colors color, char pos) : Piece(Types::Rook, color, pos, {}) {}
+
+    virtual bool movePattern(char pos) const override {
+        if (m_pos == INVALID_POS) return false;
+        auto cur = coord();
+        auto next = positions[pos];
+        return cur[R] == next[R] || cur[F] == next[F];
+    }
+
+    virtual void reinit(const State& state) override;
+};
+
+struct Queen : public Piece {
+    Queen(Colors color, char pos) : Piece(Types::Queen, color, pos, {}) {}
+
+    virtual bool movePattern(char pos) const override {
+        if (m_pos == INVALID_POS) return false;
+        auto cur = coord();
+        auto next = positions[pos];
+        return cur[R] == next[R] || cur[F] == next[F] || cur[R] - cur[F] == next[R] - next[F] || cur[R] + cur[F] == next[R] + next[F];
+    }
+
+    virtual void reinit(const State& state) override;
+};
+
+struct King : public Piece {
+    King(Colors color, char pos) : Piece(Types::King, color, pos, {}) {}
+
+    virtual bool movePattern(char pos) const override {
+        if (m_pos == INVALID_POS) return false;
+        auto cur = coord();
+        auto next = positions[pos];
+        Direction diff = {char(next[R] - cur[R]), char(next[F] - cur[F])};
+        return diff[R]*diff[R] + diff[F]*diff[F] <= 2;
+    }
+
+    virtual void reinit(const State& state) override;
+};
+
+struct PieceSet {
+    Piece* begin() { return &p1; }
+    Piece* end() { return &r2 + 1; }
+    const Piece* begin() const { return &p1; }
+    const Piece* end() const { return &r2 + 1; }
+    Piece& operator[](int i) { return *(begin() + i); }
+    const Piece& operator[](int i) const { return *(begin() + i); }
+
+    Pawn   p1;
+    Pawn   p2;
+    Pawn   p3;
+    Pawn   p4;
+    Pawn   p5;
+    Pawn   p6;
+    Pawn   p7;
+    Pawn   p8;
+    Rook   r1;
+    Knight n1;
+    Bishop b1;
+    Queen  q;
+    King   k;
+    Bishop b2;
+    Knight n2;
+    Rook   r2;
+};
+
+struct State {
+    State();
+    PieceSet blacks;
+    PieceSet whites;
+    Board board;
+    Pins blackPins;
+    Pins whitePins;
+};
+
+Direction findPin(const Piece& piece, const State& state) {
+    auto& pins = piece.color() ? state.blackPins : state.whitePins;
+    auto it = std::find_if(pins.begin(), pins.end(), [&] (const Pin& pin) { return pin.pinned == &piece; });
+    if (it != pins.end()) return it->d;
+    return {0, 0};
+}
+
+struct Find {
+    Find(const Board& board) : m_board(board) {}
+    bool operator() (char pos) const { return m_board[pos]; }
+    const Board& m_board;
+};
+
+struct Add {
+    Add(const Board& board, std::set<char>& moves, Piece::Colors color) : m_board(board), m_moves(moves), m_color(color) {}
+    bool operator() (char pos) const {
+        if (!m_board[pos] || m_board[pos]->color() != m_color) m_moves.insert(pos);
+        return m_board[pos];
+    }
+    const Board& m_board;
+    std::set<char>& m_moves;
+    Piece::Colors m_color;
+};
+
+void Pawn::reinit(const State& state) {
+    if (m_pos == INVALID_POS) return;
+    if (!m_update) return;
+    m_update = false;
+    m_allowed.clear();
+
+    auto pin = findPin(*this, state);
+
+    auto & left = m_color ? SW : NW;
+    auto & right = m_color ? SE : NE;
+
+    for (auto& direction : filter(pin, { left, right })) {
+        auto pos = makeStep(m_pos, direction);
+        if (pos != INVALID_POS && state.board[pos] && state.board[pos]->color() != m_color) m_allowed.insert(pos);
+    }
+
+    auto & forward = m_color ? S : N;
+    if (!filter(pin, {forward}).empty()) {
+        traverse(m_pos, forward, [&] (char pos) {
+                if (!state.board[pos]) m_allowed.insert(pos);
+                return state.board[pos] || !is_first_move();
+            }, 2);
+    }
+}
+
+void Knight::reinit(const State& state) {
+    if (m_pos == INVALID_POS) return;
+    if (!m_update) return;
+    m_update = false;
+    m_allowed.clear();
+    auto pin = findPin(*this, state);
+    if (pin[R] != 0 || pin[F] != 0) return;
+    for (auto& direction : { NNE, ENE, ESE, SSE, SSW, WSW, WNW, NNW }) {
+        auto pos = makeStep(m_pos, direction);
+        if (pos != INVALID_POS && (!state.board[pos] || state.board[pos]->color() != m_color)) m_allowed.insert(pos);
+    }
+}
+
+void Bishop::reinit(const State& state) {
+    if (m_pos == INVALID_POS) return;
+    if (!m_update) return;
+    m_update = false;
+    m_allowed.clear();
+    auto pin = findPin(*this, state);
+    for (auto& direction : filter(pin, { NE, SE, SW, NW })) {
+        traverse(m_pos, direction, Add(state.board, m_allowed, m_color));
+    }
+}
+
+void Rook::reinit(const State& state) {
+    if (m_pos == INVALID_POS) return;
+    if (!m_update) return;
+    m_update = false;
+    m_allowed.clear();
+    auto pin = findPin(*this, state);
+    for (auto& direction : filter(pin, { N, E, S, W })) {
+        traverse(m_pos, direction, Add(state.board, m_allowed, m_color));
+    }
+}
+
+void Queen::reinit(const State& state) {
+    if (m_pos == INVALID_POS) return;
+    if (!m_update) return;
+    m_update = false;
+    m_allowed.clear();
+    auto pin = findPin(*this, state);
+    for (auto& direction : filter(pin, { N, NE, E, SE, S, SW, W, NW })) {
+        traverse(m_pos, direction, Add(state.board, m_allowed, m_color));
+    }
+}
+
+void King::reinit(const State& state) {
+    if (m_pos == INVALID_POS) return;
+    if (!m_update) return;
+    m_update = false;
+    m_allowed.clear();
+    auto& enemyPieces = m_color ? state.whites : state.blacks;
+    auto& pawnAttackLeft = m_color ? SW : NW;
+    auto& pawnAttackRight = m_color ? SE : NE;
+    for (auto& direction : { N, NE, E, SE, S, SW, W, NW }) {
+        auto pos = makeStep(m_pos, direction);
+        bool accept = pos != INVALID_POS && !(state.board[pos] && state.board[pos]->color() == m_color);
+        if (accept) {
+            for (auto& p : enemyPieces) {
+                if (!p.movePattern(pos)) continue;
+                if (p.type() == Piece::Knight || p.type() == Piece::King) {
+                    accept = false;
+                    break;
+                }
+                else if (p.type() == Piece::Pawn) {
+                    auto from = positions[pos];
+                    auto to = p.coord();
+                    Direction d {char(to[R] - from[R]), char(to[F] - from[F])};
+                    if (d == pawnAttackLeft || d == pawnAttackRight) {
+                        accept = false;
+                        break;
+                    }
+                }
+                else {
+                    auto from = positions[pos];
+                    auto to = p.coord();
+                    Direction d = normalize({char(to[R] - from[R]), char(to[F] - from[F])});
+                    auto reached = traverse(pos, d, Find(state.board));
+                    if (p.pos() == reached) {
+                        accept = false;
+                        break;
+                    }
+                }
+            }
+        }
+        if (accept) m_allowed.insert(pos);
+    }
+}
+
+const char* Piece::name() const {
+    static_assert(pieceNames.size() == Piece::NUM_PIECES, "Mismatch between piece names and types");
+    return pieceNames[m_type];
+}
+
+char Piece::initial() const {
+    static_assert(blackShort.size() == Piece::NUM_PIECES, "Mismatch between piece names and types");
+    static_assert(whiteShort.size() == Piece::NUM_PIECES, "Mismatch between piece names and types");
+    return m_color ? blackShort[m_type] : whiteShort[m_type];
+}
+
+void Piece::invalidate() {
+    m_update = true;
+}
+
+
+const char* Piece::coord() const {
+    if (m_pos == INVALID_POS) return "";
+    return positions[m_pos];
+}
+
+bool Piece::canReach(char pos) const {
+    return movePattern(pos) && m_allowed.count(pos);
+}
+
+void Piece::take() {
+    m_pos = INVALID_POS;
+    m_allowed = {};
+}
+
+State::State()
+    : blacks {
+        {Piece::Black, "a7"_P, {"a5"_P, "a6"_P} },
+        {Piece::Black, "b7"_P, {"b5"_P, "b6"_P} },
+        {Piece::Black, "c7"_P, {"c5"_P, "c6"_P} },
+        {Piece::Black, "d7"_P, {"d5"_P, "d6"_P} },
+        {Piece::Black, "e7"_P, {"e5"_P, "e6"_P} },
+        {Piece::Black, "f7"_P, {"f5"_P, "f6"_P} },
+        {Piece::Black, "g7"_P, {"g5"_P, "g6"_P} },
+        {Piece::Black, "h7"_P, {"h5"_P, "h6"_P} },
+        {Piece::Black, "a8"_P},
+        {Piece::Black, "b8"_P, {"a6"_P, "c6"_P} },
+        {Piece::Black, "c8"_P},
+        {Piece::Black, "d8"_P},
+        {Piece::Black, "e8"_P},
+        {Piece::Black, "f8"_P},
+        {Piece::Black, "g8"_P, {"f6"_P, "h6"_P} },
+        {Piece::Black, "h8"_P},
+    }
+    , whites {
+        {Piece::White, "a2"_P, {"a3"_P, "a4"_P} },
+        {Piece::White, "b2"_P, {"b3"_P, "b4"_P} },
+        {Piece::White, "c2"_P, {"c3"_P, "c4"_P} },
+        {Piece::White, "d2"_P, {"d3"_P, "d4"_P} },
+        {Piece::White, "e2"_P, {"e3"_P, "e4"_P} },
+        {Piece::White, "f2"_P, {"f3"_P, "f4"_P} },
+        {Piece::White, "g2"_P, {"g3"_P, "g4"_P} },
+        {Piece::White, "h2"_P, {"h3"_P, "h4"_P} },
+        {Piece::White, "a1"_P},
+        {Piece::White, "b1"_P, {"a3"_P, "c3"_P} },
+        {Piece::White, "c1"_P},
+        {Piece::White, "d1"_P},
+        {Piece::White, "e1"_P},
+        {Piece::White, "f1"_P},
+        {Piece::White, "g1"_P, {"f3"_P, "h3"_P} },
+        {Piece::White, "h1"_P},
+    }
+    , board {{
+        &whites[ 8],  &whites[ 9],  &whites[10],  &whites[11],  &whites[12],  &whites[13],  &whites[14],  &whites[15],
+        &whites[ 0],  &whites[ 1],  &whites[ 2],  &whites[ 3],  &whites[ 4],  &whites[ 5],  &whites[ 6],  &whites[ 7],
+        nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,
+        nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,
+        nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,
+        nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,      nullptr,
+        &blacks[ 0],  &blacks[ 1],  &blacks[ 2],  &blacks[ 3],  &blacks[ 4],  &blacks[ 5],  &blacks[ 6],  &blacks[ 7],
+        &blacks[ 8],  &blacks[ 9],  &blacks[10],  &blacks[11],  &blacks[12],  &blacks[13],  &blacks[14],  &blacks[15],
+    }}
+{}
+
+Chessboard::Chessboard()
+    : m_state(new State())
+{
+    setGrammar();
+}
+
+Chessboard::~Chessboard() = default;
+
+void Chessboard::setPrompt(const std::string& prompt) {
+    m_prompt = prompt;
+    setGrammar();
+}
+
+void Chessboard::setGrammar() {
+    m_grammar.clear();
+
+    std::string result;
+    if (m_prompt.empty()) {
+        result += "move ::= \" \" ((piece | frompos) \" \" \"to \"?)? topos\n";
+        //result += "move ::= \" \" frompos \" \" \"to \"? topos\n";
+    }
+    else {
+        // result += "move ::= prompt \" \" ((piece | frompos) \" \" \"to \"?)? topos\n"
+        result += "move ::= prompt \" \" frompos \" \" \"to \"? topos\n"
+        "prompt ::= \" " + m_prompt + "\"\n";
+    }
+
+    std::set<Piece::Types> pieceTypes;
+    std::set<char> from_pos;
+    std::set<char> to_pos;
+    auto& pieces =  m_moveCounter % 2 ? m_state->blacks : m_state->whites;
+    std::set<size_t> flags;
+    for (auto& p : pieces) {
+        if (p.allowed().empty()) continue;
+        bool addPiece = false;
+        if (!m_inCheck || p.type() == Piece::King) {
+            to_pos.insert(p.allowed().begin(), p.allowed().end());
+            addPiece = !p.allowed().empty();
+        }
+        else {
+            for (auto move : p.allowed()) {
+                if (m_allowedInCheck.count(move)) {
+                    to_pos.insert(move);
+                    addPiece = true;
+                }
+            }
+        }
+        if (addPiece) {
+            pieceTypes.insert(p.type());
+            from_pos.insert(p.pos());
+        }
+    }
+    if (pieceTypes.empty()) return;
+
+    result += "piece ::= (";
+    for (auto& p : pieceTypes) result += " \"" + std::string(pieceNames[p]) + "\" |";
+    result.pop_back();
+    result += ")\n\n";
+
+    result += "frompos ::= (";
+    for (auto& p : from_pos) result += " \"" + std::string(positions[p]) + "\" |";
+    result.pop_back();
+    result += ")\n";
+
+    result += "topos ::= (";
+    for (auto& p : to_pos) result += " \"" + std::string(positions[p]) + "\" |";
+    result.pop_back();
+    result += ")\n";
+
+    m_grammar = std::move(result);
+}
+
+std::string Chessboard::stringifyBoard() {
+    std::string result;
+    result.reserve(16 + 2 * 64 + 16);
+    for (char rank = 'a'; rank <= 'h'; ++rank) {
+        result.push_back(rank);
+        result.push_back(' ');
+    }
+    result.back() = '\n';
+    for (int i = 7; i >= 0; --i) {
+        for (int j = 0; j < 8; ++j) {
+            auto p = m_state->board[i * 8 + j];
+            if (p) result.push_back(p->initial());
+            else result.push_back((i + j) % 2 ? '.' : '*');
+            result.push_back(' ');
+        }
+        result.push_back('0' + i + 1);
+        result.push_back('\n');
+    }
+    return result;
+}
+
+std::string Chessboard::process(const std::string& command) {
+    const auto t_start = std::chrono::high_resolution_clock::now();
+    auto color = Piece::Colors(m_moveCounter % 2);
+    Piece* piece = nullptr;
+    auto pos_to = INVALID_POS;
+    if (!parseCommand(command, piece, pos_to)) return "";
+
+    auto pos_from = piece->pos();
+
+    if (!move(*piece, pos_to)) return "";
+
+    flagUpdates(pos_from, pos_to);
+
+    detectChecks();
+
+    auto& enemyPieces = color ? m_state->whites : m_state->blacks;
+    for (auto& p : enemyPieces) p.reinit(*m_state); // only enemy moves needed next
+
+    std::string result = {positions[pos_from][R], positions[pos_from][F], '-', positions[pos_to][R], positions[pos_to][F]};
+    ++m_moveCounter;
+    setGrammar();
+    const auto t_end = std::chrono::high_resolution_clock::now();
+    auto t_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t_end - t_start).count();
+    fprintf(stdout, "%s: Move '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", result.data(), "\033[0m", (int) t_ms);
+    if (m_grammar.empty()) result.push_back('#');
+    return result;
+}
+
+bool Chessboard::parseCommand(const std::string& command, Piece*& piece, char& pos_to) {
+    auto color = Piece::Colors(m_moveCounter % 2);
+    fprintf(stdout, "%s: Command to %s: '%s%.*s%s'\n", __func__, (color ? "Black" : "White"), "\033[1m", int(command.size()), command.data(), "\033[0m");
+
+    if (command.empty()) return false;
+    auto tokens = split(command, ' ');
+    auto pos_from = INVALID_POS;
+    auto type = Piece::Types::NUM_PIECES;
+    if (tokens.size() == 1) {
+        type = Piece::Types::Pawn;
+        pos_to = strToPos(tokens.front());
+    }
+    else {
+        pos_from = strToPos(tokens.front());
+        if (pos_from == INVALID_POS) type = Piece::Types(strToType(tokens.front()));
+        pos_to = strToPos(tokens.back());
+    }
+    if (pos_to == INVALID_POS) return false;
+    if (pos_from == INVALID_POS) {
+        if (type == Piece::Types::NUM_PIECES) return false;
+        auto& pieces = color ? m_state->blacks : m_state->whites;
+        for (auto& p : pieces) {
+            if (p.type() == type && p.canReach(pos_to)) {
+                pos_from = p.pos();
+                break;
+            }
+        }
+    }
+    if (pos_from == INVALID_POS) return false;
+    if (m_state->board[pos_from] == nullptr) return false;
+    piece = m_state->board[pos_from];
+    if (piece->color() != color) return false;
+    return true;
+}
+
+void Chessboard::flagUpdates(char pos_from, char pos_to) {
+    auto color = Piece::Colors(m_moveCounter % 2);
+    auto& enemyPieces = color ? m_state->whites : m_state->blacks;
+    auto& ownPieces = color ? m_state->blacks : m_state->whites;
+    for (auto& p : enemyPieces) {
+        if (p.movePattern(pos_to) || p.movePattern(pos_from)) {
+            updatePins(p);
+            p.invalidate();
+        }
+    }
+
+    for (auto& p : ownPieces) {
+        if (p.movePattern(pos_to) || p.movePattern(pos_from)) {
+            updatePins(p);
+            p.invalidate();
+        }
+    }
+}
+
+void Chessboard::updatePins(Piece& piece) {
+    if (piece.type() == Piece::Pawn || piece.type() == Piece::Knight || piece.type() == Piece::King) return;
+    auto& enemyPieces = piece.color() ? m_state->whites : m_state->blacks;
+    auto& enemyPins = piece.color() ? m_state->whitePins : m_state->blackPins;
+    auto& king = enemyPieces.k;
+    auto it = std::find_if(enemyPins.begin(), enemyPins.end(), [&] (const Pin& pin) { return pin.pinner == &piece; });
+    if (it != enemyPins.end()) {
+        it->pinned->invalidate();
+        enemyPins.erase(it);
+    }
+    if (piece.movePattern(king.pos())) {
+        auto to = positions[king.pos()];
+        auto from = piece.coord();
+        Direction d = normalize({char(to[R] - from[R]), char(to[F] - from[F])});
+
+        auto reached = traverse(piece.pos(), d, Find(m_state->board));
+        auto foundPiece = m_state->board[reached];
+        if (&king == foundPiece) {
+            // check
+            king.invalidate();
+        }
+        else if (foundPiece && foundPiece->color() != piece.color()) {
+            reached = traverse(reached, d, Find(m_state->board));
+            if (&king == m_state->board[reached]) {
+                enemyPins.push_back({d, &piece, foundPiece});
+                foundPiece->invalidate();
+            }
+        }
+    }
+}
+
+void Chessboard::detectChecks() {
+    auto color = Piece::Colors(m_moveCounter % 2);
+    auto& enemyPieces = color ? m_state->whites : m_state->blacks;
+    auto& ownPieces = color ? m_state->blacks : m_state->whites;
+    auto& king = enemyPieces.k;
+    auto& pawnAttackLeft = color ? SW : NW;
+    auto& pawnAttackRight = color ? SE : NE;
+    for (auto& p : ownPieces) {
+        if (!p.movePattern(king.pos())) continue;
+        auto to = positions[king.pos()];
+        auto from = p.coord();
+
+        if (p.type() == Piece::Knight) {
+            if (!m_inCheck) {
+                m_allowedInCheck = { p.pos() };
+            }
+            else {
+                m_allowedInCheck.clear();
+            }
+            m_inCheck = true;
+        }
+        else if (p.type() == Piece::Pawn) {
+            Direction d {char(to[R] - from[R]), char(to[F] - from[F])};
+            if (d == pawnAttackLeft || d == pawnAttackRight) {
+                if (!m_inCheck) {
+                    m_allowedInCheck = { p.pos() };
+                }
+                else {
+                    m_allowedInCheck.clear();
+                }
+                m_inCheck = true;
+            }
+        }
+        else {
+            Direction d = normalize({char(to[R] - from[R]), char(to[F] - from[F])});
+            std::set<char> tmp;
+            auto pos = traverse(p.pos(), d, Add(m_state->board, tmp, king.color()));
+            if (pos == king.pos()) {
+                tmp.insert(p.pos());
+                if (!m_inCheck) {
+                    m_allowedInCheck = std::move(tmp);
+                }
+                else {
+                    m_allowedInCheck.clear();
+                }
+                m_inCheck = true;
+            }
+        }
+    }
+}
+
+bool Chessboard::move(Piece& piece, char pos_to) {
+    auto& allowed = piece.allowed();
+
+    if (allowed.count(pos_to) == 0 || (m_inCheck && piece.type() != Piece::King && m_allowedInCheck.count(pos_to) == 0)) return false;
+    if (m_state->board[pos_to] && m_state->board[pos_to]->color() == piece.color()) return false;
+    if (m_state->board[pos_to]) m_state->board[pos_to]->take();
+    m_state->board[piece.pos()] = nullptr;
+    m_state->board[pos_to] = &piece;
+    piece.setPos(pos_to);
+
+    m_inCheck = false;
+    m_allowedInCheck.clear();
+
+    return true;
+}

examples/wchess/libwchess/Chessboard.h

@@ -0,0 +1,33 @@
+#pragma once
+#include <string>
+#include <set>
+#include <memory>
+
+// just basic validation
+// fixme: missing en passant, castling, promotion, etc.
+struct State;
+class Piece;
+class Chessboard {
+public:
+    Chessboard();
+    ~Chessboard();
+    std::string process(const std::string& command);
+    std::string stringifyBoard();
+    const std::string& grammar() { return m_grammar; }
+    const std::string& prompt() { return m_prompt; }
+    void setPrompt(const std::string& prompt);
+private:
+    bool parseCommand(const std::string& command, Piece*& piece, char& pos_to);
+    bool move(Piece& piece, char pos);
+    void flagUpdates(char pos_from, char pos_to);
+    void updatePins(Piece& piece);
+    void detectChecks();
+    void setGrammar();
+
+    std::unique_ptr<State> m_state;
+    std::set<char> m_allowedInCheck;
+    bool m_inCheck = false;
+    int m_moveCounter = 0;
+    std::string m_grammar;
+    std::string m_prompt;
+};

examples/wchess/libwchess/WChess.cpp

@@ -0,0 +1,193 @@
+#include "WChess.h"
+#include "Chessboard.h"
+#include "grammar-parser.h"
+#include "common.h"
+#include <thread>
+
+WChess::WChess(whisper_context * ctx,
+        const whisper_full_params & wparams,
+        callbacks cb,
+        settings s)
+        : m_ctx(ctx)
+        , m_wparams(wparams)
+        , m_cb(cb)
+        , m_settings(s)
+        , m_board(new Chessboard())
+{}
+
+WChess::~WChess() = default;
+
+void WChess::set_move(const std::string& moves, float prob) const {
+    if (m_cb.set_move) (*m_cb.set_move)(moves, prob);
+}
+
+void WChess::set_grammar(const std::string& grammar) const {
+    if (m_cb.set_grammar) (*m_cb.set_grammar)(grammar);
+}
+
+bool WChess::get_audio(std::vector<float>& pcmf32) const {
+    if (m_cb.get_audio) return (*m_cb.get_audio)(pcmf32);
+    return false;
+}
+
+std::string WChess::stringify_board() const {
+    return m_board->stringifyBoard();
+}
+
+std::string WChess::get_grammar() const {
+    return m_board->grammar();
+}
+
+void WChess::run() {
+    bool have_prompt  = true;
+    bool ask_prompt   = !have_prompt;
+
+    float logprob_min  = 0.0f;
+
+    float logprob_sum  = 0.0f;
+
+    int n_tokens  = 0;
+
+    std::vector<float> pcmf32_cur;
+    std::vector<float> pcmf32_prompt;
+
+    const std::string k_prompt = have_prompt ? "" : "rook to d4, f3";
+    int64_t t_ms = 0;
+
+    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;
+    }
+
+    while (get_audio(pcmf32_cur)) {
+        if (!pcmf32_cur.empty()) {
+            // fprintf(stdout, "%s: Processing ...\n", __func__);
+
+            if (!have_prompt) {
+                const auto txt = ::trim(transcribe(pcmf32_cur, logprob_min, logprob_sum, n_tokens, 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;
+                    m_board->setPrompt(k_prompt);
+                }
+            } else {
+                if (!pcmf32_prompt.empty()) pcmf32_cur.insert(pcmf32_cur.begin(), pcmf32_prompt.begin(), pcmf32_prompt.end());
+                constexpr size_t MIN_SIZE = 1.2 * WHISPER_SAMPLE_RATE;
+                if (MIN_SIZE > pcmf32_cur.size()) pcmf32_cur.insert(pcmf32_cur.begin(), MIN_SIZE - pcmf32_cur.size(), 0.0f);
+
+                // fprintf(stdout, "%s: grammar rules:\n'%s'\n", __func__, m_board->grammar().c_str());
+
+                auto grammar_parsed = grammar_parser::parse(m_board->grammar().c_str());
+                auto grammar_rules  = grammar_parsed.c_rules();
+
+                m_wparams.grammar_rules   = grammar_rules.data();
+                m_wparams.n_grammar_rules = grammar_rules.size();
+
+                m_wparams.i_start_rule    = grammar_parsed.symbol_ids.at("move");
+                auto txt = ::trim(transcribe(pcmf32_cur, logprob_min, logprob_sum, n_tokens, t_ms));
+
+                const float p = 100.0f * std::exp(logprob_min);
+
+                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;
+                    }
+                }
+
+                fprintf(stdout, "%s:   DEBUG: txt = '%s', prob = %.2f%%\n", __func__, txt.c_str(), p);
+                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");
+
+                if (!command.empty()) {
+                    set_move(m_board->process(command), p);
+                    set_grammar(m_board->grammar());
+                }
+                if (m_board->grammar().empty()) {
+                    fprintf(stdout, "%s: No more moves possible\n", __func__);
+                    break;
+                }
+            }
+        }
+
+        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;
+        }
+    }
+}
+
+std::string WChess::transcribe(
+                const std::vector<float> & pcmf32,
+                float & logprob_min,
+                float & logprob_sum,
+                int & n_tokens,
+                int64_t & t_ms) {
+    const auto t_start = std::chrono::high_resolution_clock::now();
+
+    logprob_min = 0.0f;
+    logprob_sum = 0.0f;
+    n_tokens    = 0;
+    t_ms = 0;
+
+    if (whisper_full(m_ctx, m_wparams, pcmf32.data(), pcmf32.size()) != 0) {
+        return {};
+    }
+
+    std::string result;
+
+    const int n_segments = whisper_full_n_segments(m_ctx);
+    for (int i = 0; i < n_segments; ++i) {
+        const char * text = whisper_full_get_segment_text(m_ctx, i);
+
+        result += text;
+
+        const int n = whisper_full_n_tokens(m_ctx, i);
+        for (int j = 0; j < n; ++j) {
+            const auto token = whisper_full_get_token_data(m_ctx, i, j);
+
+            if(token.plog > 0.0f) return {};
+            logprob_min = std::min(logprob_min, token.plog);
+            logprob_sum += token.plog;
+            ++n_tokens;
+        }
+    }
+
+    const auto t_end = std::chrono::high_resolution_clock::now();
+    t_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t_end - t_start).count();
+
+    return result;
+}

examples/wchess/libwchess/WChess.h

@@ -0,0 +1,63 @@
+#pragma once
+#include "whisper.h"
+#include <string>
+#include <vector>
+#include <memory>
+
+class Chessboard;
+
+class WChess {
+public:
+    using CheckRunningCb = bool (*)();
+    using GetAudioCb = bool (*)(std::vector<float> &);
+    using SetMovesCb = void (*)(const std::string &, float);
+    using SetGrammarCb = void (*)(const std::string &);
+    using ClearAudioCb = void (*)();
+
+    struct callbacks {
+        GetAudioCb get_audio = nullptr;
+        SetMovesCb set_move = nullptr;
+        SetGrammarCb set_grammar = nullptr;
+    };
+
+    struct settings {
+        int32_t vad_ms     = 2000;
+        int32_t prompt_ms  = 5000;
+        int32_t command_ms = 4000;
+        float vad_thold    = 0.2f;
+        float freq_thold   = 100.0f;
+        bool print_energy  = false;
+    };
+
+    WChess(
+        whisper_context * ctx,
+        const whisper_full_params & wparams,
+        callbacks cb,
+        settings s
+    );
+    ~WChess();
+
+    void run();
+
+    std::string stringify_board() const;
+
+    std::string get_grammar() const;
+
+private:
+    bool get_audio(std::vector<float>& pcmf32) const;
+    void set_move(const std::string& moves, float prob) const;
+    void set_grammar(const std::string& grammar) const;
+
+    std::string transcribe(
+                    const std::vector<float> & pcmf32,
+                    float & logprob_min,
+                    float & logprob_sum,
+                    int & n_tokens,
+                    int64_t & t_ms);
+
+    whisper_context * m_ctx;
+    whisper_full_params m_wparams;
+    const callbacks m_cb;
+    const settings m_settings;
+    std::unique_ptr<Chessboard> m_board;
+};

examples/wchess/libwchess/test-chessboard.cpp

@@ -0,0 +1,117 @@
+#include "Chessboard.h"
+
+#define ASSERT(x) \
+    do { \
+        if (!(x)) { \
+            fprintf(stderr, "ASSERT: %s:%d: %s\n", __FILE__, __LINE__, #x); \
+            fflush(stderr); \
+            exit(1); \
+        } \
+    } while (0)
+
+
+int main() {
+    {
+        Chessboard chess;
+
+        ASSERT(chess.process("pawn to d4") == "d2-d4");
+        ASSERT(chess.process("e5") == "e7-e5");
+        ASSERT(chess.process("c1 h6") == "c1-h6");
+        ASSERT(chess.process("queen h4") == "d8-h4");
+        ASSERT(chess.process("bishop to g5") == "h6-g5");
+        ASSERT(chess.process("bishop to b4") == "f8-b4");
+        ASSERT(chess.process("c4") == "");
+        ASSERT(chess.process("knight c3") == "b1-c3");
+        ASSERT(chess.process("knight c6") == "b8-c6");
+        ASSERT(chess.process("f3") == "");
+    }
+
+    {
+        Chessboard chess;
+
+        ASSERT(chess.process("d4") == "d2-d4");
+        ASSERT(chess.process("e5") == "e7-e5");
+        ASSERT(chess.process("e4") == "e2-e4");
+        ASSERT(chess.process("queen h4") == "d8-h4");
+        ASSERT(chess.process("queen h5") == "d1-h5");
+        ASSERT(chess.process("f5") == "");
+        ASSERT(chess.process("g6") == "g7-g6");
+        ASSERT(chess.process("knight e2") == "g1-e2");
+        ASSERT(chess.process("f5") == "f7-f5");
+        ASSERT(chess.process("knight g3") == "e2-g3");
+        ASSERT(chess.process("g5") == "");
+        ASSERT(chess.process("king e7") == "e8-e7");
+        ASSERT(chess.process("f4") == "f2-f4");
+        ASSERT(chess.process("g5") == "g6-g5");
+    }
+
+    {
+        Chessboard chess;
+
+        ASSERT(chess.process("e4") == "e2-e4");
+        ASSERT(chess.process("c5") == "c7-c5");
+        ASSERT(chess.process("e5") == "e4-e5");
+        ASSERT(chess.process("c4") == "c5-c4");
+        ASSERT(chess.process("e6") == "e5-e6");
+        ASSERT(chess.process("c3") == "c4-c3");
+        ASSERT(chess.process("e7") == "");
+        ASSERT(chess.process("f7") == "e6-f7");
+        ASSERT(chess.process("d2") == "");
+        ASSERT(chess.process("king to f7") == "e8-f7");
+        ASSERT(chess.process("f4") == "f2-f4");
+        ASSERT(chess.process("d2") == "c3-d2");
+        ASSERT(chess.process("f5") == "");
+        ASSERT(chess.process("king to e2") == "e1-e2");
+        ASSERT(chess.process("king to g6") == "f7-g6");
+        ASSERT(chess.process("f5") == "f4-f5");
+        ASSERT(chess.process("e6") == "");
+        ASSERT(chess.process("king to h5") == "g6-h5");
+        ASSERT(chess.process("g4") == "g2-g4");
+        ASSERT(chess.process("king to g5") == "h5-g5");
+        ASSERT(chess.process("h4") == "h2-h4");
+        ASSERT(chess.process("king to h5") == "");
+        ASSERT(chess.process("king to g6") == "");
+        ASSERT(chess.process("king to h6") == "g5-h6");
+        ASSERT(chess.process("bishop to d2") == "c1-d2");
+        ASSERT(chess.process("king to g5") == "");
+        ASSERT(chess.process("g5") == "g7-g5");
+    }
+
+    {
+        Chessboard chess;
+        ASSERT(chess.process("f4") == "f2-f4");
+        ASSERT(chess.process("e5") == "e7-e5");
+        ASSERT(chess.process("g4") == "g2-g4");
+        ASSERT(chess.process("queen to h4") == "d8-h4#");
+        ASSERT(chess.process("knight f3") == "");
+        ASSERT(chess.grammar().empty());
+    }
+
+    {
+        Chessboard chess;
+        ASSERT(chess.process("f4") == "f2-f4");
+        ASSERT(chess.process("e5") == "e7-e5");
+        ASSERT(chess.process("g4") == "g2-g4");
+        ASSERT(chess.process("d5") == "d7-d5");
+        ASSERT(chess.process("g1 f3") == "g1-f3");
+        ASSERT(chess.process("queen to h4") == "d8-h4");
+        ASSERT(!chess.grammar().empty());
+    }
+
+    {
+        Chessboard chess;
+        ASSERT(chess.process("knight c3") == "b1-c3");
+        ASSERT(chess.process("knight c6") == "b8-c6");
+        ASSERT(chess.process("knight b5") == "c3-b5");
+        ASSERT(chess.process("knight f6") == "g8-f6");
+        ASSERT(chess.process("knight d6") == "b5-d6");
+        ASSERT(chess.process("knight d4") == "");
+        ASSERT(chess.process("d6") == "c7-d6");
+        ASSERT(chess.process("e4") == "e2-e4");
+        ASSERT(chess.process("knight d4") == "c6-d4");
+        ASSERT(chess.process("d3") == "d2-d3");
+        ASSERT(chess.process("knight e4") == "f6-e4");
+        ASSERT(chess.process("king to e2") == "");
+        ASSERT(chess.process("king to d2") == "");
+    }
+}

examples/wchess/wchess.cmd/CMakeLists.txt

@@ -0,0 +1,8 @@
+if (WHISPER_SDL2)
+    set(TARGET wchess)
+    add_executable(${TARGET} wchess.cmd.cpp)
+
+    include(DefaultTargetOptions)
+
+    target_link_libraries(${TARGET} PRIVATE wchess-core common-sdl ${CMAKE_THREAD_LIBS_INIT})
+endif ()

examples/wchess/wchess.cmd/wchess.cmd.cpp

@@ -0,0 +1,247 @@
+// Command line voice assisted chess
+//
+// Speak chess move commands to the microphone.
+// The moves will translated to chessboard positions.
+//
+//
+
+#include "WChess.h"
+#include "common-sdl.h"
+#include <iostream>
+
+#include <memory>
+#include <thread>
+
+// command-line parameters
+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 = 8000;
+    int32_t capture_id = -1;
+    int32_t max_tokens = 32;
+    int32_t audio_ctx  = 0;
+
+    float vad_thold  = 0.6f;
+    float freq_thold = 100.0f;
+
+    float grammar_penalty = 100.0f;
+
+    bool speed_up      = false;
+    bool translate     = false;
+    bool print_special = false;
+    bool print_energy  = false;
+    bool no_timestamps = true;
+    bool use_gpu       = true;
+
+    std::string language  = "en";
+    std::string model     = "models/ggml-base.en.bin";
+    std::string fname_out;
+    std::string commands;
+    std::string prompt;
+    std::string context;
+    std::string grammar;
+};
+
+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");
+    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, "  -pms N,     --prompt-ms N    [%-7d] prompt duration in milliseconds\n",             params.prompt_ms);
+    fprintf(stderr, "  -cms N,     --command-ms N   [%-7d] command duration in milliseconds\n",            params.command_ms);
+    fprintf(stderr, "  -c ID,      --capture ID     [%-7d] capture device ID\n",                           params.capture_id);
+    fprintf(stderr, "  -mt N,      --max-tokens N   [%-7d] maximum number of tokens per audio chunk\n",    params.max_tokens);
+    fprintf(stderr, "  -ac N,      --audio-ctx N    [%-7d] audio context size (0 - all)\n",                params.audio_ctx);
+    fprintf(stderr, "  -vth N,     --vad-thold N    [%-7.2f] voice activity detection threshold\n",        params.vad_thold);
+    fprintf(stderr, "  -fth N,     --freq-thold N   [%-7.2f] high-pass frequency cutoff\n",                params.freq_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, "  -ps,        --print-special  [%-7s] print special tokens\n",                        params.print_special ? "true" : "false");
+    fprintf(stderr, "  -pe,        --print-energy   [%-7s] print sound energy (for debugging)\n",          params.print_energy ? "true" : "false");
+    fprintf(stderr, "  -ng,        --no-gpu         [%-7s] disable GPU\n",                                 params.use_gpu ? "false" : "true");
+    fprintf(stderr, "  -l LANG,    --language LANG  [%-7s] spoken language\n",                             params.language.c_str());
+    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, "  -ctx,       --context        [%-7s] sample text to help the transcription\n",       params.context.c_str());
+    fprintf(stderr, "  --grammar-penalty N          [%-7.1f] scales down logits of nongrammar tokens\n",   params.grammar_penalty);
+    fprintf(stderr, "\n");
+}
+
+bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
+    for (int i = 1; i < argc; i++) {
+        std::string arg = argv[i];
+
+        if (arg == "-h" || arg == "--help") {
+            whisper_print_usage(argc, argv, params);
+            exit(0);
+        }
+        else if (arg == "-t"   || arg == "--threads")       { params.n_threads     = std::stoi(argv[++i]); }
+        else if (arg == "-pms" || arg == "--prompt-ms")     { params.prompt_ms     = std::stoi(argv[++i]); }
+        else if (arg == "-cms" || arg == "--command-ms")    { params.command_ms    = std::stoi(argv[++i]); }
+        else if (arg == "-c"   || arg == "--capture")       { params.capture_id    = std::stoi(argv[++i]); }
+        else if (arg == "-mt"  || arg == "--max-tokens")    { params.max_tokens    = std::stoi(argv[++i]); }
+        else if (arg == "-ac"  || arg == "--audio-ctx")     { params.audio_ctx     = std::stoi(argv[++i]); }
+        else if (arg == "-vth" || arg == "--vad-thold")     { params.vad_thold     = std::stof(argv[++i]); }
+        else if (arg == "-fth" || arg == "--freq-thold")    { params.freq_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 == "-ps"  || arg == "--print-special") { params.print_special = true; }
+        else if (arg == "-pe"  || arg == "--print-energy")  { params.print_energy  = true; }
+        else if (arg == "-ng"  || arg == "--no-gpu")        { params.use_gpu       = false; }
+        else if (arg == "-l"   || arg == "--language")      { params.language      = argv[++i]; }
+        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 if (arg == "-ctx" || arg == "--context")       { params.context       = argv[++i]; }
+        else if (                 arg == "--grammar-penalty") { params.grammar_penalty = std::stof(argv[++i]); }
+        else {
+            fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
+            whisper_print_usage(argc, argv, params);
+            exit(0);
+        }
+    }
+
+    return true;
+}
+
+std::unique_ptr<WChess> g_wchess;
+int g_moveCount = 0;
+void set_move(const std::string & move, float) {
+    if (!move.empty()) {
+        g_moveCount++;
+        fprintf(stdout, "Move: %s\n\n", move.c_str());
+    }
+    else fprintf(stdout, "Move rejected\n\n");
+    fprintf(stdout, "%s\n", g_wchess->stringify_board().c_str());
+    fprintf(stdout, "%s\n", g_moveCount ? "White's turn" : "Black's turn");
+}
+
+audio_async g_audio(30*1000);
+bool g_listening = false;
+std::vector<float> g_pcmf32;
+
+bool read_input() {
+    std::string input;
+    while (true) {
+        fprintf(stdout, "[(l)isten/(p)ause/(q)uit]: ");
+        std::cin >> input;
+        fprintf(stdout, "\n");
+        if (input[0] == 'q') {
+            fprintf(stdout, "Quitting\n");
+            return false;
+        }
+        if (input[0] == 'l') {
+            if (!g_listening) {
+                fprintf(stdout, "Listening\n");
+                g_listening = true;
+                g_pcmf32.clear();
+                g_audio.resume();
+                g_audio.clear();
+            }
+            else fprintf(stdout, "Still listening\n");
+            return true;
+        }
+        else {
+            if (g_listening) {
+                g_listening = false;
+                g_audio.get(0, g_pcmf32);
+                g_audio.pause();
+                fprintf(stdout, "Processing\n");
+            }
+            else fprintf(stdout, "Not listening\n");
+            return true;
+        }
+    }
+    return true;
+}
+
+bool get_audio(std::vector<float> & pcmf32_cur) {
+    if (!read_input()) return false;
+    if (!g_pcmf32.empty()) pcmf32_cur = std::move(g_pcmf32);
+    else pcmf32_cur.clear();
+    return true;
+}
+
+int main(int argc, char ** argv) {
+    whisper_params params;
+
+    if (whisper_params_parse(argc, argv, params) == false) {
+        return 1;
+    }
+
+    if (whisper_lang_id(params.language.c_str()) == -1) {
+        fprintf(stderr, "error: unknown language '%s'\n", params.language.c_str());
+        whisper_print_usage(argc, argv, params);
+        exit(0);
+    }
+
+    // whisper init
+
+    struct whisper_context_params cparams;
+    cparams.use_gpu = params.use_gpu;
+
+    struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+    if (!ctx) {
+        fprintf(stderr, "%s: whisper_init_from_file_with_params() failed!\n", __func__);
+        return 1;
+    }
+
+    // init audio
+
+    if (!g_audio.init(params.capture_id, WHISPER_SAMPLE_RATE)) {
+        fprintf(stderr, "%s: audio.init() failed!\n", __func__);
+        return 1;
+    }
+
+    struct whisper_full_params wparams = whisper_full_default_params(whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY);
+    wparams.offset_ms        = 0;
+    wparams.translate        = false;
+    wparams.no_context       = true;
+    wparams.single_segment   = true;
+    wparams.print_realtime   = false;
+    wparams.print_progress   = false;
+    wparams.print_timestamps = true;
+    wparams.print_special    = false;
+    wparams.no_timestamps    = true;
+
+    wparams.max_tokens       = 32;
+    wparams.audio_ctx        = 768; // partial encoder context for better performance
+
+    wparams.temperature     = 0.0f;
+    wparams.temperature_inc = 2.0f;
+    wparams.greedy.best_of  = 1;
+
+    wparams.beam_search.beam_size = 1;
+
+    wparams.language         = "en";
+
+    wparams.grammar_penalty = 100.0;
+
+    wparams.initial_prompt = params.context.data();
+
+    WChess::callbacks cb;
+    cb.get_audio = get_audio;
+    cb.set_move = set_move;
+
+    WChess::settings s;
+    s.vad_ms = 2000;
+    s.prompt_ms = params.prompt_ms;
+    s.command_ms = params.command_ms;
+    s.vad_thold = params.vad_thold;
+    s.freq_thold = params.freq_thold;
+    s.print_energy = params.print_energy;
+
+    g_wchess.reset(new WChess(ctx, wparams, cb, s));
+    set_move("start", 0);
+    g_wchess->run();
+
+    whisper_print_timings(ctx);
+    whisper_free(ctx);
+
+    return 0;
+}

examples/wchess/wchess.wasm/CMakeLists.txt

@@ -0,0 +1,51 @@
+set(TARGET wchess.wasm)
+
+add_executable(${TARGET}
+    wchess.wasm.cpp
+    )
+
+include(DefaultTargetOptions)
+
+target_link_libraries(${TARGET} PRIVATE
+    common
+    wchess-core
+    )
+
+unset(EXTRA_FLAGS)
+
+if (WHISPER_WASM_SINGLE_FILE)
+    set(EXTRA_FLAGS "-s SINGLE_FILE=1")
+    message(STATUS "Embedding WASM inside chess.js")
+
+    add_custom_command(
+        TARGET ${TARGET} POST_BUILD
+        COMMAND ${CMAKE_COMMAND} -E copy
+        ${CMAKE_BINARY_DIR}/bin/${TARGET}.js
+        ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/js/chess.js
+        )
+endif()
+
+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 FORCE_FILESYSTEM=1 \
+    -s EXPORTED_RUNTIME_METHODS=\"['print', 'printErr', 'ccall', 'cwrap']\" \
+    ${EXTRA_FLAGS} \
+    ")
+
+
+add_custom_command(
+        TARGET ${TARGET} POST_BUILD
+        COMMAND ${CMAKE_COMMAND} -E copy_directory
+        ${CMAKE_CURRENT_SOURCE_DIR}/chessboardjs-1.0.0
+        ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/
+        COMMAND ${CMAKE_COMMAND} -E copy
+        ${CMAKE_CURRENT_SOURCE_DIR}/jquery-3.7.1.min.js
+        ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/js/
+    )
+
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/index-tmpl.html  ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/index.html @ONLY)
+configure_file(${CMAKE_SOURCE_DIR}/examples/helpers.js    ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/js/helpers.js @ONLY)

examples/wchess/wchess.wasm/chessboardjs-1.0.0/css/chessboard-1.0.0.css

@@ -0,0 +1,54 @@
+/*! chessboard.js v1.0.0 | (c) 2019 Chris Oakman | MIT License chessboardjs.com/license */
+
+.clearfix-7da63 {
+  clear: both;
+}
+
+.board-b72b1 {
+  border: 2px solid #404040;
+  box-sizing: content-box;
+}
+
+.square-55d63 {
+  float: left;
+  position: relative;
+
+  /* disable any native browser highlighting */
+  -webkit-touch-callout: none;
+    -webkit-user-select: none;
+     -khtml-user-select: none;
+       -moz-user-select: none;
+        -ms-user-select: none;
+            user-select: none;
+}
+
+.white-1e1d7 {
+  background-color: #f0d9b5;
+  color: #b58863;
+}
+
+.black-3c85d {
+  background-color: #b58863;
+  color: #f0d9b5;
+}
+
+.highlight1-32417, .highlight2-9c5d2 {
+  box-shadow: inset 0 0 3px 3px yellow;
+}
+
+.notation-322f9 {
+  cursor: default;
+  font-family: "Helvetica Neue", Helvetica, Arial, sans-serif;
+  font-size: 14px;
+  position: absolute;
+}
+
+.alpha-d2270 {
+  bottom: 1px;
+  right: 3px;
+}
+
+.numeric-fc462 {
+  top: 2px;
+  left: 2px;
+}

examples/wchess/wchess.wasm/chessboardjs-1.0.0/css/chessboard-1.0.0.min.css

@@ -0,0 +1,2 @@
+/*! chessboard.js v1.0.0 | (c) 2019 Chris Oakman | MIT License chessboardjs.com/license */
+.clearfix-7da63{clear:both}.board-b72b1{border:2px solid #404040;box-sizing:content-box}.square-55d63{float:left;position:relative;-webkit-touch-callout:none;-webkit-user-select:none;-khtml-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none}.white-1e1d7{background-color:#f0d9b5;color:#b58863}.black-3c85d{background-color:#b58863;color:#f0d9b5}.highlight1-32417,.highlight2-9c5d2{box-shadow:inset 0 0 3px 3px #ff0}.notation-322f9{cursor:default;font-family:"Helvetica Neue",Helvetica,Arial,sans-serif;font-size:14px;position:absolute}.alpha-d2270{bottom:1px;right:3px}.numeric-fc462{top:2px;left:2px}

examples/wchess/wchess.wasm/chessboardjs-1.0.0/js/chessboard-1.0.0/CHANGELOG.md

@@ -0,0 +1,32 @@
+# chessboard.js Change Log
+
+All notable changes to this project will be documented in this file.
+
+## [1.0.0] - 2019-06-11
+- Orientation methods now return current orientation. [Issue #64]
+- Drop support for IE8
+- Do not check for `window.JSON` (Error #1004)
+- Rename `ChessBoard` to `Chessboard` (`ChessBoard` is still supported, however)
+- id query selectors are now supported as the first argument to `Chessboard()`
+- Remove Error #1002
+- Format code according to [StandardJS]
+- Bump minimum jQuery version to 1.8.3
+- Throttle piece drag functions
+
+## [0.3.0] - 2013-08-10
+- Added `appearSpeed` animation config property
+- Added `onSnapbackEnd` event
+- Added `onMoveEnd` event
+
+## [0.2.0] - 2013-08-05
+- Added `onMouseoverSquare` and `onMouseoutSquare` events
+- Added `onSnapEnd` event
+- Added square code as CSS class on the squares
+- Added [chess.js] integration examples
+
+## [0.1.0] - 2013-05-21
+- Initial release
+
+[chess.js]:https://github.com/jhlywa/chess.js
+[Issue #64]:https://github.com/oakmac/chessboardjs/issues/64
+[StandardJS]:https://standardjs.com/

examples/wchess/wchess.wasm/chessboardjs-1.0.0/js/chessboard-1.0.0/LICENSE.md

@@ -0,0 +1,20 @@
+Copyright 2019 Chris Oakman
+
+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.

examples/wchess/wchess.wasm/chessboardjs-1.0.0/js/chessboard-1.0.0/README.md

@@ -0,0 +1,82 @@
+# chessboard.js
+
+chessboard.js is a JavaScript chessboard component. It depends on [jQuery].
+
+Please see [chessboardjs.com] for documentation and examples.
+
+## What is chessboard.js?
+
+chessboard.js is a JavaScript chessboard component with a flexible "just a
+board" API that
+
+chessboard.js is a standalone JavaScript Chess Board. It is designed to be "just
+a board" and expose a powerful API so that it can be used in different ways.
+Here's a non-exhaustive list of things you can do with chessboard.js:
+
+- Use chessboard.js to show game positions alongside your expert commentary.
+- Use chessboard.js to have a tactics website where users have to guess the best
+  move.
+- Integrate chessboard.js and [chess.js] with a PGN database and allow people to
+  search and playback games (see [Example 5000])
+- Build a chess server and have users play their games out using the
+  chessboard.js board.
+
+chessboard.js is flexible enough to handle any of these situations with relative
+ease.
+
+## What can chessboard.js **not** do?
+
+The scope of chessboard.js is limited to "just a board." This is intentional and
+makes chessboard.js flexible for handling a multitude of chess-related problems.
+
+This is a common source of confusion for new users. [remove?]
+
+Specifically, chessboard.js does not understand anything about how the game of
+chess is played: how a knight moves, who's turn is it, is White in check?, etc.
+
+Fortunately, the powerful [chess.js] library deals with exactly this sort of
+problem domain and plays nicely with chessboard.js's flexible API. Some examples
+of chessboard.js combined with chess.js: 5000, 5001, 5002
+
+Please see the powerful [chess.js] library for an API to deal with these sorts
+of questions.
+
+
+This logic is distinct from the logic of the board. Please see the powerful
+[chess.js] library for this aspect of your application.
+
+
+
+Here is a list of things that chessboard.js is **not**:
+
+- A chess engine
+- A legal move validator
+- A PGN parser
+
+chessboard.js is designed to work well with any of those things, but the idea
+behind chessboard.js is that the logic that controls the board should be
+independent of those other problems.
+
+## Docs and Examples
+
+- Docs - <http://chessboardjs.com/docs>
+- Examples - <http://chessboardjs.com/examples>
+
+## Developer Tools
+
+```sh
+# create a build in the build/ directory
+npm run build
+
+# re-build the website
+npm run website
+```
+
+## License
+
+[MIT License](LICENSE.md)
+
+[jQuery]:https://jquery.com/
+[chessboardjs.com]:http://chessboardjs.com
+[chess.js]:https://github.com/jhlywa/chess.js
+[Example 5000]:http://chessboardjs.com/examples#5000

examples/wchess/wchess.wasm/chessboardjs-1.0.0/js/chessboard-1.0.0/package.json

@@ -0,0 +1,29 @@
+{
+  "author": "Chris Oakman <chris@oakmac.com> (http://chrisoakman.com/)",
+  "name": "@chrisoakman/chessboardjs",
+  "description": "JavaScript chessboard widget",
+  "homepage": "https://chessboardjs.com",
+  "license": "MIT",
+  "version": "1.0.0",
+  "repository": {
+    "type": "git",
+    "url": "git://github.com/oakmac/chessboardjs.git"
+  },
+  "files": ["dist/"],
+  "dependencies": {
+    "jquery": ">=3.4.1"
+  },
+  "devDependencies": {
+    "csso": "3.5.1",
+    "fs-plus": "3.1.1",
+    "kidif": "1.1.0",
+    "mustache": "2.3.0",
+    "standard": "10.0.2",
+    "uglify-js": "3.6.0"
+  },
+  "scripts": {
+    "build": "standard lib/chessboard.js && node scripts/build.js",
+    "standard": "standard --fix lib/*.js website/js/*.js",
+    "website": "node scripts/website.js"
+  }
+}

examples/wchess/wchess.wasm/index-tmpl.html

@@ -0,0 +1,499 @@
+<!doctype html>
+<html lang="en-us">
+    <head>
+        <title>wchess : voice-controlled chess using Whisper + WebAssembly</title>
+        <script src="https://cdnjs.cloudflare.com/ajax/libs/iframe-resizer/4.3.1/iframeResizer.contentWindow.min.js"></script>
+
+        <meta name="viewport" content="width=device-width, initial-scale=0.7, maximum-scale=1, minimum-scale=0.7, user-scalable=no"/>
+        <meta name="apple-mobile-web-app-capable" content="yes" />
+
+        <style>
+            #output {
+                width: 100%;
+                height: 100%;
+                margin: 0 auto;
+                margin-top: 10px;
+                border-left: 0px;
+                border-right: 0px;
+                padding-left: 0px;
+                padding-right: 0px;
+                display: block;
+                background-color: black;
+                color: white;
+                font-size: 10px;
+                font-family: 'Lucida Console', Monaco, monospace;
+                outline: none;
+                white-space: pre;
+                overflow-wrap: normal;
+                overflow-x: scroll;
+            }
+            .button {
+                background-color: #000000;
+                color: #FFFFFF;
+                padding: 20px;
+                border-radius: 10px;
+                -moz-border-radius: 10px;
+                -webkit-border-radius: 10px;
+                margin:10px;
+                width:  100px;
+                height:  50px;
+                -webkit-touch-callout: none; /* Safari */
+                -webkit-user-select: none; /* Chrome */
+                -moz-user-select: none; /* Firefox */
+                -ms-user-select: none; /* Internet Explorer/Edge */
+                user-select: none;
+            }
+            button[disabled]{
+                background-color: #cccccc;
+                color: #666666;
+                padding: 20px;
+                border-radius: 10px;
+                -moz-border-radius: 10px;
+                -webkit-border-radius: 10px;
+                margin:10px;
+                width: 100px;
+            }
+            .center {
+                display: flex;
+                justify-content: center;
+                align-items: center;
+                width: 500px;
+            }
+            #description {
+                width: 500px;
+            }
+        </style>
+        <link rel="stylesheet" href="css/chessboard-1.0.0.min.css" integrity="sha384-q94+BZtLrkL1/ohfjR8c6L+A6qzNH9R2hBLwyoAfu3i/WCvQjzL2RQJ3uNHDISdU" crossorigin="anonymous">
+    </head>
+    <body>
+        <div id="main-container">
+            <div id="description">
+                <b>wchess : voice-controlled chess using Whisper + WebAssembly</b>
+
+                <br><br>
+
+                This is a demonstration of using Whisper to recognize voice commands in the browser.
+
+                <br><br>
+
+                Usage:<br>
+
+                <ul>
+                    <li>Select a Whisper model</li>
+                    <li>Accept the microphone permission request if prompted</li>
+                    <li>Hold the button and say a chess move (e.g. "Knight to c3")</li>
+                    <li>Release the button and wait for the move to be recognized</li>
+                    <li>Repeat</li>
+                </ul>
+
+                Examples:<br>
+
+                <ul>
+                    <li><b>"d4"</b></li>
+                    <li><b>"e2 e4"</b></li>
+                    <li><b>"Knight f3"</b></li>
+                    <li><b>"Bishop to b5"</b></li>
+                </ul>
+
+                Features:<br>
+
+                <ul>
+                    <li>Model quantization for reduced memory footprint (~42MB)</li>
+                    <li><a href="https://github.com/ggerganov/whisper.cpp/pull/1229">Grammar-based sampling</a> for improved recognition accuracy</li>
+                </ul>
+
+                <b>
+                Note that not all chess moves are supported. For example, castling and pawn promotion
+                currently do not work, but can be easily implemented. There could also be some bugs in
+                the move handling logic in general. The main reason for that is to keep the implementation
+                simple. The assumption is that a real application would already have a proper move
+                validation logic in place.<br><br>
+
+                The main purpose of this example is to demonstrate the capabilities of whisper.cpp and
+                its application in the browser for voice recognition locally on your device.
+                </b>
+
+                <br><br>
+
+                You can find more about this project on <a href="https://github.com/ggerganov/whisper.cpp/tree/master/examples/wchess">GitHub</a>.
+
+                <br><br>
+
+                <b>More examples:</b>
+                    <a href="https://whisper.ggerganov.com/">main</a> |
+                    <a href="https://whisper.ggerganov.com/bench">bench</a> |
+                    <a href="https://whisper.ggerganov.com/stream">stream</a> |
+                    <a href="https://whisper.ggerganov.com/command">command</a> |
+                    <a href="https://whisper.ggerganov.com/talk">talk</a> |
+
+                <br><br>
+
+            </div>
+
+            <hr>
+
+            <div id="model-whisper">
+                Whisper model: <span id="model-whisper-status"></span>
+                <button id="fetch-whisper-tiny-en" onclick="loadWhisper()">tiny.en (Q8_0, 42 MB)</button>
+                <span id="fetch-whisper-progress"></span>
+                <br><br>
+                <button id="clear" onclick="clearCache()">Clear browser cache</button>
+                <!--
+                    <input type="file" id="file" name="file" onchange="loadFile(event, 'whisper.bin')" />
+                -->
+            </div>
+
+            <div id="game">
+                <br>
+                <div id="chessboard" style="width: 500px"></div>
+                <script src="js/jquery-3.7.1.min.js"></script>
+                <script src="js/chessboard-1.0.0.min.js"></script>
+                <script>
+                    var board = Chessboard('chessboard', 'start')
+                    var move_count = 0;
+                </script>
+
+                <br>
+
+                <div id="state">
+                    Status: <b><span id="state-status">select model</span></b>
+
+                    <div id="input" class="center">
+                        <button id="toggler" class="button" onselectstart="return false" style="display: none">Hold</button>
+                    </div>
+
+                    <pre id="state-grammar">[The grammar will be displayed here]</pre>
+
+                    <pre id="state-moves">[The moves will be displayed here]</pre>
+                </div>
+            </div>
+
+            <hr>
+
+            Debug output:
+            <textarea id="output" rows="20"></textarea>
+
+            <br>
+
+            <b>Troubleshooting</b>
+
+            <br><br>
+
+            The page does some heavy computations, so make sure:
+
+            <ul>
+                <li>To use a modern web browser (e.g. Chrome, Firefox)</li>
+                <li>Your browser supports WASM <a href="https://webassembly.org/roadmap/">Fixed-width SIMD</a></li>
+            </ul>
+
+            <div class="cell-version">
+                <span>
+                    |
+                    Build time: <span class="nav-link">@GIT_DATE@</span> |
+                    Commit hash: <a class="nav-link" href="https://github.com/ggerganov/whisper.cpp/commit/@GIT_SHA1@">@GIT_SHA1@</a> |
+                    Commit subject: <span class="nav-link">@GIT_COMMIT_SUBJECT@</span> |
+                    <a class="nav-link" href="https://github.com/ggerganov/whisper.cpp/tree/master/examples/command.wasm">Source Code</a> |
+                </span>
+            </div>
+        </div>
+
+        <script type="text/javascript" src="js/helpers.js"></script>
+        <script type='text/javascript'>
+            // web audio context
+            var context = null;
+
+            // the command instance
+            var instance = null;
+
+            // model name
+            var model_whisper = null;
+            var model_file = null;
+
+            var module_ready = null;
+
+            var Module = {
+                print: printTextarea,
+                printErr: printTextarea,
+                setStatus: function(text) {
+                    printTextarea('js: ' + text);
+                },
+                monitorRunDependencies: function(left) {
+                },
+                preRun: function() {
+                    printTextarea('js: Preparing ...');
+                },
+                postRun: function() {
+                    printTextarea('js: Module initialized successfully!');
+                    module_ready = true;
+                    initInstance();
+                }
+            };
+
+            function initInstance() {
+                if (!module_ready || !model_file || instance) return
+
+                instance = Module.init(model_file);
+
+                if (instance) {
+                    setStatus('Ready');
+                    printTextarea("js: whisper initialized, instance: " + instance);
+                }
+                else {
+                    printTextarea("js: failed to initialize whisper");
+                }
+            }
+
+            function setStatus(text) {
+                document.getElementById('state-status').innerHTML = text;
+            }
+
+            //
+            // fetch models
+            //
+
+            let dbVersion = 1
+            let dbName    = 'whisper.ggerganov.com';
+            let indexedDB = window.indexedDB || window.mozIndexedDB || window.webkitIndexedDB || window.msIndexedDB
+
+            function storeFS(fname, buf) {
+                // write to WASM file using FS_createDataFile
+                // if the file exists, delete it
+                try {
+                    Module.FS_unlink(fname);
+                } catch (e) {
+                    // ignore
+                }
+
+                Module.FS_createDataFile("/", fname, buf, true, true);
+
+                printTextarea('storeFS: stored model: ' + fname + ' size: ' + buf.length);
+
+                document.getElementById('model-whisper-status').innerHTML = 'loaded "' + model_whisper + '"!';
+
+                model_file = fname;
+                initInstance();
+            }
+
+            function loadWhisper() {
+                setStatus('Loading')
+                //let url     = 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en-q8_0.bin';
+                let url     = 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en-q8_0.bin';
+                let dst     = 'whisper.bin';
+                let size_mb = 42;
+
+                model_whisper = 'tiny.en-q8_0';
+
+                document.getElementById('model-whisper-status').innerHTML = 'loading "' + model_whisper + '" ... ';
+                document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
+
+                cbProgress = function(p) {
+                    let el = document.getElementById('fetch-whisper-progress');
+                    el.innerHTML = Math.round(100*p) + '%';
+                };
+
+                cbCancel = function() {
+                    var el;
+                    el = document.getElementById('model-whisper-status');  if (el) el.innerHTML = '';
+                };
+
+                loadRemote(url, dst, size_mb, cbProgress, storeFS, cbCancel, printTextarea);
+
+                // init audio capture so that the user receives a permission request
+                {
+                    let context = new AudioContext({
+                        sampleRate: 16000,
+                        channelCount: 1,
+                        echoCancellation: false,
+                        autoGainControl:  true,
+                        noiseSuppression: true,
+                    });
+                    navigator.mediaDevices.getUserMedia({audio: true, video: false})
+                        .then(function(s) {
+                            stream = s;
+                            stream.getTracks().forEach(function(track) {
+                                track.stop();
+                            });
+                        })
+                        .catch(function(err) {
+                            printTextarea('js: error getting audio stream: ' + err);
+                        });
+                    context.close();
+                }
+
+                document.getElementById('toggler').style.display = 'block';
+            }
+
+            //
+            // microphone
+            //
+
+            const kSampleRate = 16000;
+            const kRestartRecording_s = 120;
+            const kIntervalAudio_ms = 250; // pass the recorded audio to the C++ instance at this rate
+
+            var mediaRecorder = null;
+            var doRecording = false;
+            var startTime = 0;
+
+            window.AudioContext = window.AudioContext || window.webkitAudioContext;
+            window.OfflineAudioContext = window.OfflineAudioContext || window.webkitOfflineAudioContext;
+
+            function stopRecording() {
+                if (mediaRecorder) {
+                    mediaRecorder.stop();
+                }
+            }
+
+            function startRecording() {
+                if (!context) {
+                    context = new AudioContext({
+                        sampleRate: kSampleRate,
+                        channelCount: 1,
+                        echoCancellation: false,
+                        autoGainControl:  true,
+                        noiseSuppression: true,
+                    });
+                }
+
+                startTime = Date.now();
+
+                var chunks = [];
+                var stream = null;
+
+                navigator.mediaDevices.getUserMedia({audio: true, video: false})
+                    .then(function(s) {
+                        stream = s;
+                        mediaRecorder = new MediaRecorder(stream);
+                        mediaRecorder.ondataavailable = function(e) {
+                            chunks.push(e.data);
+
+                            var blob = new Blob(chunks, { 'type' : 'audio/ogg; codecs=opus' });
+                            var reader = new FileReader();
+
+                            reader.onload = function(event) {
+                                var buf = new Uint8Array(reader.result);
+                                context.decodeAudioData(buf.buffer, function(audioBuffer) {
+                                    var offlineContext = new OfflineAudioContext(audioBuffer.numberOfChannels, audioBuffer.length, audioBuffer.sampleRate);
+                                    var source = offlineContext.createBufferSource();
+                                    source.buffer = audioBuffer;
+                                    source.connect(offlineContext.destination);
+                                    source.start(0);
+
+                                    offlineContext.startRendering().then(function(renderedBuffer) {
+                                        let audio = renderedBuffer.getChannelData(0);
+                                        printTextarea('js: number of samples: ' + audio.length);
+                                        Module.set_audio(instance, audio);
+                                    });
+
+                                    mediaRecorder = null;
+                                    context = null;
+                                });
+                            }
+
+                            reader.readAsArrayBuffer(blob);
+                        };
+
+                        mediaRecorder.onstop = function(e) {
+                            stream.getTracks().forEach(function(track) {
+                                track.stop();
+                            });
+                        };
+
+                        mediaRecorder.start();
+                    })
+                    .catch(function(err) {
+                        printTextarea('js: error getting audio stream: ' + err);
+                    });
+            }
+
+            //
+            // main
+            //
+
+            var nLines = 0;
+            var movesAll = '';
+
+            // document.body.addEventListener('keydown', function(event) {
+            //     if (event.keyCode === 32) {
+            //         document.getElementById('toggler').innerText = "";
+            //         onStart();
+            //     }
+            // }, true);
+
+            // document.body.addEventListener('keyup', function(event) {
+            //     if (event.keyCode === 32) {
+            //         document.getElementById('toggler').innerText = "Hold";
+            //         onStop();
+            //     }
+            // }, true);
+
+            document.getElementById('toggler').addEventListener("touchstart", function(event){
+                this.innerText = "";
+                onStart();
+            }, true);
+
+            document.getElementById('toggler').addEventListener("touchend", function(event){
+                this.innerText = "Hold";
+                onStop();
+            }, true)
+
+            document.getElementById('toggler').addEventListener('mousedown', function(event) {
+                this.innerText = "";
+                onStart();
+            }, true);
+
+            document.getElementById('toggler').addEventListener('mouseup', function(event) {
+                this.innerText = "Hold";
+                onStop();
+            }, true);
+
+            function onStart() {
+                if (!instance) return;
+                setStatus('Listening');
+
+                startRecording();
+            }
+
+            function onStop() {
+                setStatus('Processing');
+                printTextarea('js: stopping recording ...');
+                stopRecording();
+            }
+
+            function setMove(move, prob) {
+                if (move != null && move.length > 1) {
+                    let gameOver =  move[move.length - 1] === '#';
+                    if (gameOver) {
+                        move = move.substring(0, move.length - 1);
+                        document.getElementById('toggler').disabled = true;
+                    }
+                    board.move(move);
+
+                    movesAll += move + ', prob = ' + prob.toFixed(2) + '% <br>';
+                    nLines++;
+
+                    // if more than 10 lines, remove the first line
+                    if (nLines > 10) {
+                        var i = movesAll.indexOf('<br>');
+                        if (i > 0) {
+                            movesAll = movesAll.substring(i + 4);
+                            nLines--;
+                        }
+                    }
+                    ++move_count;
+                    setStatus(gameOver ? 'Done' : move_count % 2 ? 'Black\'s turn' : 'White\'s turn');
+                    document.getElementById('state-moves').innerHTML = movesAll;
+                }
+                else {
+                    setStatus('Failed. ' + (move_count % 2 ? 'Black\'s turn' : 'White\'s turn'));
+                }
+            }
+
+            function setGrammar(grammar) {
+                document.getElementById('state-grammar').innerHTML = grammar;
+            }
+
+        </script>
+        <script type="text/javascript" src="js/chess.js"></script>
+    </body>
+</html>

examples/wchess/wchess.wasm/wchess.wasm.cpp

@@ -0,0 +1,141 @@
+#include <WChess.h>
+#include <emscripten.h>
+#include <emscripten/bind.h>
+
+#include <thread>
+
+constexpr int N_THREAD = 8;
+
+std::vector<struct whisper_context *> g_contexts(4, nullptr);
+
+std::mutex  g_mutex;
+std::thread g_worker;
+
+std::condition_variable g_cv;
+
+bool g_running(false);
+std::vector<float> g_pcmf32;
+
+void set_move(const std::string & move, float prob) {
+    MAIN_THREAD_EM_ASM({
+        setMove(UTF8ToString($0), $1)
+    }, move.c_str(), prob);
+}
+
+void set_grammar(const std::string & grammar) {
+    MAIN_THREAD_EM_ASM({
+        setGrammar(UTF8ToString($0))
+    }, grammar.c_str());
+}
+
+bool get_audio(std::vector<float> & audio) {
+    std::unique_lock<std::mutex> lock(g_mutex);
+    g_cv.wait(lock, [] { return !g_running || !g_pcmf32.empty(); });
+    if (!g_running) return false;
+    audio = std::move(g_pcmf32);
+    return true;
+}
+
+void wchess_main(size_t i) {
+    struct whisper_full_params wparams = whisper_full_default_params(whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY);
+
+    wparams.n_threads        = std::min(N_THREAD, (int) std::thread::hardware_concurrency());
+    wparams.offset_ms        = 0;
+    wparams.translate        = false;
+    wparams.no_context       = true;
+    wparams.single_segment   = true;
+    wparams.print_realtime   = false;
+    wparams.print_progress   = false;
+    wparams.print_timestamps = true;
+    wparams.print_special    = false;
+    wparams.no_timestamps    = true;
+
+    wparams.max_tokens       = 32;
+    wparams.audio_ctx        = 1280; // partial encoder context for better performance
+
+    wparams.temperature      = 0.0f;
+    wparams.temperature_inc  = 2.0f;
+    wparams.greedy.best_of   = 1;
+
+    wparams.beam_search.beam_size = 1;
+
+    wparams.language         = "en";
+
+    wparams.grammar_penalty = 100.0;
+    wparams.initial_prompt = "bishop to c3, rook to d4, knight to e5, d4 d5, knight to c3, c3, queen to d4, king b1, pawn to a1, bishop to b2, knight to c3,";
+
+    printf("command: using %d threads\n", wparams.n_threads);
+
+    WChess::callbacks cb;
+    cb.get_audio = get_audio;
+    cb.set_move = set_move;
+    cb.set_grammar = set_grammar;
+
+    WChess(g_contexts[i], wparams, cb, {}).run();
+
+    if (i < g_contexts.size()) {
+        whisper_free(g_contexts[i]);
+        g_contexts[i] = nullptr;
+    }
+}
+
+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_from_file_with_params(path_model.c_str(), whisper_context_default_params());
+                if (g_contexts[i] != nullptr) {
+                    g_running = true;
+                    if (g_worker.joinable()) {
+                        g_worker.join();
+                    }
+                    g_worker = std::thread([i]() {
+                        wchess_main(i);
+                    });
+
+                    return i + 1;
+                } else {
+                    return (size_t) 0;
+                }
+            }
+        }
+
+        return (size_t) 0;
+    }));
+
+    emscripten::function("free", emscripten::optional_override([](size_t /* index */) {
+        {
+            std::unique_lock<std::mutex> lock(g_mutex);
+            g_running = false;
+        }
+        g_cv.notify_one();
+    }));
+
+    emscripten::function("set_audio", emscripten::optional_override([](size_t index, const emscripten::val & audio) {
+        --index;
+
+        if (index >= g_contexts.size()) {
+            return -1;
+        }
+
+        if (g_contexts[index] == nullptr) {
+            return -2;
+        }
+
+        {
+            std::lock_guard<std::mutex> lock(g_mutex);
+            const int n = audio["length"].as<int>();
+
+            emscripten::val heap = emscripten::val::module_property("HEAPU8");
+            emscripten::val memory = heap["buffer"];
+
+            g_pcmf32.resize(n);
+
+            emscripten::val memoryView = audio["constructor"].new_(memory, reinterpret_cast<uintptr_t>(g_pcmf32.data()), n);
+            memoryView.call<void>("set", audio);
+        }
+        g_cv.notify_one();
+
+        return 0;
+    }));
+}

examples/whisper.android.java/.gitignore

@@ -0,0 +1,15 @@
+*.iml
+.gradle
+/local.properties
+/.idea/caches
+/.idea/libraries
+/.idea/modules.xml
+/.idea/workspace.xml
+/.idea/navEditor.xml
+/.idea/assetWizardSettings.xml
+.DS_Store
+/build
+/captures
+.externalNativeBuild
+.cxx
+local.properties

examples/whisper.android.java/README.md

@@ -0,0 +1,20 @@
+A sample Android app using java code and [whisper.cpp](https://github.com/ggerganov/whisper.cpp/) to do voice-to-text transcriptions.
+
+To use:
+
+1. Select a model from the [whisper.cpp repository](https://github.com/ggerganov/whisper.cpp/tree/master/models).[^1]
+2. Copy the model to the "app/src/main/assets/models" folder.
+3. Select a sample audio file (for example, [jfk.wav](https://github.com/ggerganov/whisper.cpp/raw/master/samples/jfk.wav)).
+4. Copy the sample to the "app/src/main/assets/samples" folder.
+5. Modify the modelFilePath in the WhisperService.java
+6. Modify the sampleFilePath in the WhisperService.java
+7. Select the "release" active build variant, and use Android Studio to run and deploy to your device.
+[^1]: I recommend the tiny or base models for running on an Android device.
+
+PS:  
+1. Do not move this android project folder individually to other folders, because this android project folder depends on the files of the whole project.  
+2. The cpp code is compiled during the build process  
+3. If you want to import a compiled cpp project in your Android project, please refer to the https://github.com/litongjava/whisper.cpp.android.java.demo  
+
+![](README_files/1.jpg)
+

examples/whisper.android.java/app/.gitignore

@@ -0,0 +1 @@
+/build

examples/whisper.android.java/app/build.gradle

@@ -0,0 +1,58 @@
+plugins {
+  id 'com.android.application'
+}
+
+android {
+  compileSdkVersion 30
+  buildToolsVersion '30.0.3'
+
+  defaultConfig {
+    applicationId "com.litongjava.whisper.android.java"
+    minSdkVersion 21
+    targetSdkVersion 30
+    versionCode 1
+    versionName "1.0"
+
+    testInstrumentationRunner "androidx.test.runner.AndroidJUnitRunner"
+    externalNativeBuild {
+      cmake {
+        cppFlags ""
+      }
+    }
+    ndk {
+      abiFilters 'arm64-v8a', 'armeabi-v7a', 'x86', 'x86_64'
+    }
+  }
+
+  buildTypes {
+    release {
+      signingConfig signingConfigs.debug
+      minifyEnabled true
+      proguardFiles getDefaultProguardFile('proguard-android-optimize.txt'), 'proguard-rules.pro'
+    }
+  }
+  externalNativeBuild {
+    cmake {
+      path "src/main/jni/whisper/CMakeLists.txt"
+    }
+  }
+  ndkVersion "25.2.9519653"
+  compileOptions {
+    sourceCompatibility JavaVersion.VERSION_1_8
+    targetCompatibility JavaVersion.VERSION_1_8
+  }
+}
+
+dependencies {
+  implementation 'androidx.appcompat:appcompat:1.1.0'
+  implementation 'com.google.android.material:material:1.1.0'
+  implementation 'androidx.constraintlayout:constraintlayout:1.1.3'
+  testImplementation 'junit:junit:4.+'
+  androidTestImplementation 'androidx.test.ext:junit:1.1.5'
+  androidTestImplementation 'androidx.test.espresso:espresso-core:3.5.1'
+
+  //litongjava
+  implementation 'com.litongjava:android-view-inject:1.0'
+  implementation 'com.litongjava:jfinal-aop:1.0.1'
+  implementation 'com.litongjava:litongjava-android-utils:1.0.0'
+}

examples/whisper.android.java/app/proguard-rules.pro

@@ -0,0 +1,21 @@
+# Add project specific ProGuard rules here.
+# You can control the set of applied configuration files using the
+# proguardFiles setting in build.gradle.
+#
+# For more details, see
+#   http://developer.android.com/guide/developing/tools/proguard.html
+
+# If your project uses WebView with JS, uncomment the following
+# and specify the fully qualified class name to the JavaScript interface
+# class:
+#-keepclassmembers class fqcn.of.javascript.interface.for.webview {
+#   public *;
+#}
+
+# Uncomment this to preserve the line number information for
+# debugging stack traces.
+#-keepattributes SourceFile,LineNumberTable
+
+# If you keep the line number information, uncomment this to
+# hide the original source file name.
+#-renamesourcefileattribute SourceFile

examples/whisper.android.java/app/src/androidTest/java/com/litongjava/whisper/android/java/ExampleInstrumentedTest.java

@@ -0,0 +1,26 @@
+package com.litongjava.whisper.android.java;
+
+import android.content.Context;
+
+import androidx.test.platform.app.InstrumentationRegistry;
+import androidx.test.ext.junit.runners.AndroidJUnit4;
+
+import org.junit.Test;
+import org.junit.runner.RunWith;
+
+import static org.junit.Assert.*;
+
+/**
+ * Instrumented test, which will execute on an Android device.
+ *
+ * @see <a href="http://d.android.com/tools/testing">Testing documentation</a>
+ */
+@RunWith(AndroidJUnit4.class)
+public class ExampleInstrumentedTest {
+  @Test
+  public void useAppContext() {
+    // Context of the app under test.
+    Context appContext = InstrumentationRegistry.getInstrumentation().getTargetContext();
+    assertEquals("com.litongjava.whisper.android.java", appContext.getPackageName());
+  }
+}

examples/whisper.android.java/app/src/main/AndroidManifest.xml

@@ -0,0 +1,22 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+  package="com.litongjava.whisper.android.java">
+
+  <application
+    android:allowBackup="true"
+    android:name=".app.App"
+    android:icon="@mipmap/ic_launcher"
+    android:label="@string/app_name"
+    android:roundIcon="@mipmap/ic_launcher_round"
+    android:supportsRtl="true"
+    android:theme="@style/Theme.Whisperandroidjava">
+    <activity android:name=".MainActivity">
+      <intent-filter>
+        <action android:name="android.intent.action.MAIN" />
+
+        <category android:name="android.intent.category.LAUNCHER" />
+      </intent-filter>
+    </activity>
+  </application>
+
+</manifest>

examples/whisper.android.java/app/src/main/assets/logback.xml

@@ -0,0 +1,40 @@
+<?xml version="1.0" encoding="UTF-8" ?>
+<configuration debug="false" xmlns="http://ch.qos.logback/xml/ns/logback"
+  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
+  xsi:schemaLocation="http://ch.qos.logback/xml/ns/logback https://raw.githubusercontent.com/enricopulatzo/logback-XSD/master/src/main/xsd/logback.xsd
+http://ch.qos.logback/xml/ns/logback ">
+  <!--Define the storage address of the log file Do not use relative paths in the LogBack configuration. -->
+  <property name="LOG_HOME" value="logs" />
+  <!--Formatted output: %d means the date, %-6level: log level from the left display 6 characters wide, %m: log message, %n is a newline character -->
+  <property name="CONSOLE_LOG_PATTERN"
+    value="%d{yyyy-MM-dd HH:mm:ss.SSS} [%thread] %-6level%logger{0}.%M:%L - %m%n" />
+
+  <!-- console output -->
+  <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
+    <encoder class="ch.qos.logback.classic.encoder.PatternLayoutEncoder">
+      <pattern>${CONSOLE_LOG_PATTERN}</pattern>
+    </encoder>
+  </appender>
+
+  <!-- Generate log files on a daily basis -->
+  <appender name="FILE" class="ch.qos.logback.core.rolling.RollingFileAppender">
+    <encoder class="ch.qos.logback.classic.encoder.PatternLayoutEncoder">
+      <pattern>${CONSOLE_LOG_PATTERN}</pattern>
+    </encoder>
+    <rollingPolicy class="ch.qos.logback.core.rolling.TimeBasedRollingPolicy">
+      <!--File name for log file output -->
+      <fileNamePattern>${LOG_HOME}/project-name-%d{yyyy-MM-dd}.log</fileNamePattern>
+      <!--Maximum size of log file -->
+      <maxHistory>180</maxHistory>
+    </rollingPolicy>
+    <!--日志文件最大的大小 -->
+    <triggeringPolicy class="ch.qos.logback.core.rolling.SizeBasedTriggeringPolicy">
+      <maxFileSize>10MB</maxFileSize>
+    </triggeringPolicy>
+  </appender>
+  <!-- Log output level and source-->
+  <root level="info">
+    <appender-ref ref="STDOUT" />
+    <appender-ref ref="FILE" />
+  </root>
+</configuration>

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/MainActivity.java

@@ -0,0 +1,107 @@
+package com.litongjava.whisper.android.java;
+
+import androidx.annotation.RequiresApi;
+import androidx.appcompat.app.AppCompatActivity;
+
+import android.content.Context;
+import android.os.Build;
+import android.os.Bundle;
+import android.os.Handler;
+import android.os.Looper;
+import android.view.View;
+import android.widget.TextView;
+
+import com.blankj.utilcode.util.ThreadUtils;
+import com.litongjava.android.view.inject.annotation.FindViewById;
+import com.litongjava.android.view.inject.annotation.FindViewByIdLayout;
+import com.litongjava.android.view.inject.annotation.OnClick;
+import com.litongjava.android.view.inject.utils.ViewInjectUtils;
+import com.litongjava.jfinal.aop.Aop;
+import com.litongjava.jfinal.aop.AopManager;
+import com.litongjava.whisper.android.java.services.WhisperService;
+import com.litongjava.whisper.android.java.task.LoadModelTask;
+import com.litongjava.whisper.android.java.task.TranscriptionTask;
+import com.litongjava.whisper.android.java.utils.AssetUtils;
+import com.whispercpp.java.whisper.WhisperLib;
+
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import java.io.File;
+
+
+@FindViewByIdLayout(R.layout.activity_main)
+public class MainActivity extends AppCompatActivity {
+
+  @FindViewById(R.id.sample_text)
+  private TextView tv;
+
+  Logger log = LoggerFactory.getLogger(this.getClass());
+  private WhisperService whisperService = Aop.get(WhisperService.class);
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  @Override
+  protected void onCreate(Bundle savedInstanceState) {
+    super.onCreate(savedInstanceState);
+    //setContentView(R.layout.activity_main);
+    ViewInjectUtils.injectActivity(this, this);
+    initAopBean();
+    showSystemInfo();
+  }
+
+  private void initAopBean() {
+    Handler mainHandler = new Handler(Looper.getMainLooper());
+    AopManager.me().addSingletonObject(mainHandler);
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  @OnClick(R.id.loadModelBtn)
+  public void loadModelBtn_OnClick(View v) {
+    Context context = getBaseContext();
+    ThreadUtils.executeByIo(new LoadModelTask(tv));
+  }
+
+  @OnClick(R.id.transcriptSampleBtn)
+  public void transcriptSampleBtn_OnClick(View v) {
+    Context context = getBaseContext();
+
+    long start = System.currentTimeMillis();
+    String sampleFilePath = "samples/jfk.wav";
+    File filesDir = context.getFilesDir();
+    File sampleFile = AssetUtils.copyFileIfNotExists(context, filesDir, sampleFilePath);
+    long end = System.currentTimeMillis();
+    String msg = "copy file:" + (end - start) + "ms";
+    outputMsg(tv, msg);
+    ThreadUtils.executeByIo(new TranscriptionTask(tv, sampleFile));
+  }
+
+  private void outputMsg(TextView tv, String msg) {
+    tv.append(msg + "\n");
+    log.info(msg);
+  }
+
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  @OnClick(R.id.systemInfoBtn)
+  public void systemInfoBtn_OnClick(View v) {
+    showSystemInfo();
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public void showSystemInfo() {
+    String systemInfo = WhisperLib.getSystemInfo();
+    tv.append(systemInfo + "\n");
+  }
+
+  @OnClick(R.id.clearBtn)
+  public void clearBtn_OnClick(View v) {
+    tv.setText("");
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  @Override
+  protected void onDestroy() {
+    super.onDestroy();
+    whisperService.release();
+  }
+}

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/app/App.java

@@ -0,0 +1,13 @@
+package com.litongjava.whisper.android.java.app;
+
+import android.app.Application;
+
+import com.blankj.utilcode.util.Utils;
+
+public class App extends Application {
+  @Override
+  public void onCreate() {
+    super.onCreate();
+    Utils.init(this);
+  }
+}

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/bean/WhisperSegment.java

@@ -0,0 +1,47 @@
+package com.litongjava.whisper.android.java.bean;
+
+/**
+ * Created by litonglinux@qq.com on 10/21/2023_7:48 AM
+ */
+public class WhisperSegment {
+  private long start, end;
+  private String sentence;
+
+  public WhisperSegment() {
+  }
+
+  public WhisperSegment(long start, long end, String sentence) {
+    this.start = start;
+    this.end = end;
+    this.sentence = sentence;
+  }
+
+  public long getStart() {
+    return start;
+  }
+
+  public long getEnd() {
+    return end;
+  }
+
+  public String getSentence() {
+    return sentence;
+  }
+
+  public void setStart(long start) {
+    this.start = start;
+  }
+
+  public void setEnd(long end) {
+    this.end = end;
+  }
+
+  public void setSentence(String sentence) {
+    this.sentence = sentence;
+  }
+
+  @Override
+  public String toString() {
+    return "["+start+" --> "+end+"]:"+sentence;
+  }
+}

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/services/WhisperService.java

@@ -0,0 +1,101 @@
+package com.litongjava.whisper.android.java.services;
+
+import android.content.Context;
+import android.os.Build;
+import android.os.Handler;
+import android.widget.TextView;
+import android.widget.Toast;
+
+import androidx.annotation.RequiresApi;
+
+import com.blankj.utilcode.util.ToastUtils;
+import com.blankj.utilcode.util.Utils;
+import com.litongjava.android.utils.dialog.AlertDialogUtils;
+import com.litongjava.jfinal.aop.Aop;
+import com.litongjava.whisper.android.java.bean.WhisperSegment;
+import com.litongjava.whisper.android.java.single.LocalWhisper;
+import com.litongjava.whisper.android.java.utils.WaveEncoder;
+
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import java.io.File;
+import java.io.IOException;
+import java.util.List;
+import java.util.concurrent.ExecutionException;
+
+public class WhisperService {
+  private Logger log = LoggerFactory.getLogger(this.getClass());
+
+  private final Object lock = new Object();
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public void loadModel(TextView tv) {
+    String modelFilePath = LocalWhisper.modelFilePath;
+    String msg = "load model from :" + modelFilePath + "\n";
+    outputMsg(tv, msg);
+
+    long start = System.currentTimeMillis();
+    LocalWhisper.INSTANCE.init();
+    long end = System.currentTimeMillis();
+    msg = "model load successful:" + (end - start) + "ms";
+    outputMsg(tv, msg);
+    ToastUtils.showLong(msg);
+
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public void transcribeSample(TextView tv, File sampleFile) {
+    String msg = "";
+    msg = "transcribe file from :" + sampleFile.getAbsolutePath();
+    outputMsg(tv, msg);
+
+    Long start = System.currentTimeMillis();
+    float[] audioData = new float[0];  // 读取音频样本
+    try {
+      audioData = WaveEncoder.decodeWaveFile(sampleFile);
+    } catch (IOException e) {
+      e.printStackTrace();
+      return;
+    }
+    long end = System.currentTimeMillis();
+    msg = "decode wave file:" + (end - start) + "ms";
+    outputMsg(tv, msg);
+
+    start = System.currentTimeMillis();
+    List<WhisperSegment> transcription = null;
+    try {
+      //transcription = LocalWhisper.INSTANCE.transcribeData(audioData);
+      transcription = LocalWhisper.INSTANCE.transcribeDataWithTime(audioData);
+    } catch (ExecutionException e) {
+      e.printStackTrace();
+    } catch (InterruptedException e) {
+      e.printStackTrace();
+    }
+    end = System.currentTimeMillis();
+    if(transcription!=null){
+      ToastUtils.showLong(transcription.toString());
+      msg = "Transcript successful:" + (end - start) + "ms";
+      outputMsg(tv, msg);
+
+      outputMsg(tv, transcription.toString());
+
+    }else{
+      msg = "Transcript failed:" + (end - start) + "ms";
+      outputMsg(tv, msg);
+    }
+
+  }
+
+  private void outputMsg(TextView tv, String msg) {
+    log.info(msg);
+    if(tv!=null){
+      Aop.get(Handler.class).post(()->{ tv.append(msg + "\n");});
+    }
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public void release() {
+    //noting to do
+  }
+}

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/single/LocalWhisper.java

@@ -0,0 +1,66 @@
+package com.litongjava.whisper.android.java.single;
+
+import android.app.Application;
+import android.os.Build;
+import android.os.Handler;
+
+import androidx.annotation.RequiresApi;
+
+import com.blankj.utilcode.util.ToastUtils;
+import com.blankj.utilcode.util.Utils;
+import com.litongjava.jfinal.aop.Aop;
+import com.litongjava.whisper.android.java.bean.WhisperSegment;
+import com.litongjava.whisper.android.java.utils.AssetUtils;
+import com.whispercpp.java.whisper.WhisperContext;
+
+import java.io.File;
+import java.util.List;
+import java.util.concurrent.ExecutionException;
+
+
+@RequiresApi(api = Build.VERSION_CODES.O)
+public enum LocalWhisper {
+  INSTANCE;
+
+  public static final String modelFilePath = "models/ggml-tiny.bin";
+  private WhisperContext whisperContext;
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  LocalWhisper() {
+    Application context = Utils.getApp();
+    File filesDir = context.getFilesDir();
+    File modelFile = AssetUtils.copyFileIfNotExists(context, filesDir, modelFilePath);
+    String realModelFilePath = modelFile.getAbsolutePath();
+    whisperContext = WhisperContext.createContextFromFile(realModelFilePath);
+  }
+
+  public synchronized String transcribeData(float[] data) throws ExecutionException, InterruptedException {
+    if(whisperContext==null){
+        toastModelLoading();
+        return null;
+    }else{
+      return whisperContext.transcribeData(data);
+    }
+  }
+
+    private static void toastModelLoading() {
+        Aop.get(Handler.class).post(()->{
+          ToastUtils.showShort("please wait for model loading");
+        });
+    }
+
+    public List<WhisperSegment> transcribeDataWithTime(float[] audioData) throws ExecutionException, InterruptedException {
+    if(whisperContext==null){
+        toastModelLoading();
+      return null;
+    }else{
+      return whisperContext.transcribeDataWithTime(audioData);
+    }
+  }
+
+  public void init() {
+    //noting to do.but init
+  }
+
+
+}

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/task/LoadModelTask.java

@@ -0,0 +1,44 @@
+package com.litongjava.whisper.android.java.task;
+
+import android.content.Context;
+import android.os.Build;
+import android.os.Handler;
+import android.widget.TextView;
+
+import com.blankj.utilcode.util.ThreadUtils;
+import com.litongjava.jfinal.aop.Aop;
+import com.litongjava.whisper.android.java.services.WhisperService;
+
+import java.io.File;
+
+public class LoadModelTask extends ThreadUtils.Task<Object> {
+  private final TextView tv;
+  public LoadModelTask(TextView tv) {
+    this.tv = tv;
+  }
+
+  @Override
+  public Object doInBackground() {
+    if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.O) {
+      Aop.get(WhisperService.class).loadModel(tv);
+    }else{
+      Aop.get(Handler.class).post(()->{
+        tv.append("not supported android devices");
+      });
+
+    }
+    return null;
+  }
+
+  @Override
+  public void onSuccess(Object result) {
+  }
+
+  @Override
+  public void onCancel() {
+  }
+
+  @Override
+  public void onFail(Throwable t) {
+  }
+}

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/task/TranscriptionTask.java

@@ -0,0 +1,44 @@
+package com.litongjava.whisper.android.java.task;
+
+import android.content.Context;
+import android.os.Build;
+import android.widget.TextView;
+
+import com.blankj.utilcode.util.ThreadUtils;
+import com.litongjava.jfinal.aop.Aop;
+import com.litongjava.whisper.android.java.services.WhisperService;
+
+import java.io.File;
+
+public class TranscriptionTask extends ThreadUtils.Task<Object> {
+  private final TextView tv;
+  private final File sampleFile;
+
+  public TranscriptionTask(TextView tv, File sampleFile) {
+    this.tv = tv;
+    this.sampleFile = sampleFile;
+
+  }
+
+  @Override
+  public Object doInBackground() {
+    if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.O) {
+      Aop.get(WhisperService.class).transcribeSample(tv, sampleFile);
+    }else{
+      tv.append("not supported android devices");
+    }
+    return null;
+  }
+
+  @Override
+  public void onSuccess(Object result) {
+  }
+
+  @Override
+  public void onCancel() {
+  }
+
+  @Override
+  public void onFail(Throwable t) {
+  }
+}

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/utils/AssetUtils.java

@@ -0,0 +1,91 @@
+package com.litongjava.whisper.android.java.utils;
+
+import android.content.Context;
+
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import java.io.BufferedInputStream;
+import java.io.BufferedOutputStream;
+import java.io.File;
+import java.io.FileNotFoundException;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.OutputStream;
+
+public class AssetUtils {
+  private static Logger log = LoggerFactory.getLogger(AssetUtils.class);
+
+  public static File copyFileIfNotExists(Context context, File distDir, String filename) {
+    File dstFile = new File(distDir, filename);
+    if (dstFile.exists()) {
+      return dstFile;
+    } else {
+      File parentFile = dstFile.getParentFile();
+      log.info("parentFile:{}", parentFile);
+      if (!parentFile.exists()) {
+        parentFile.mkdirs();
+      }
+      AssetUtils.copyFileFromAssets(context, filename, dstFile);
+    }
+    return dstFile;
+  }
+
+  public static void copyDirectoryFromAssets(Context appCtx, String srcDir, String dstDir) {
+    if (srcDir.isEmpty() || dstDir.isEmpty()) {
+      return;
+    }
+    try {
+      if (!new File(dstDir).exists()) {
+        new File(dstDir).mkdirs();
+      }
+      for (String fileName : appCtx.getAssets().list(srcDir)) {
+        String srcSubPath = srcDir + File.separator + fileName;
+        String dstSubPath = dstDir + File.separator + fileName;
+        if (new File(srcSubPath).isDirectory()) {
+          copyDirectoryFromAssets(appCtx, srcSubPath, dstSubPath);
+        } else {
+          copyFileFromAssets(appCtx, srcSubPath, dstSubPath);
+        }
+      }
+    } catch (Exception e) {
+      e.printStackTrace();
+    }
+  }
+
+  public static void copyFileFromAssets(Context appCtx, String srcPath, String dstPath) {
+    File dstFile = new File(dstPath);
+    copyFileFromAssets(appCtx, srcPath, dstFile);
+  }
+
+  public static void copyFileFromAssets(Context appCtx, String srcPath, File dstFile) {
+    if (srcPath.isEmpty()) {
+      return;
+    }
+    InputStream is = null;
+    OutputStream os = null;
+    try {
+      is = new BufferedInputStream(appCtx.getAssets().open(srcPath));
+
+      os = new BufferedOutputStream(new FileOutputStream(dstFile));
+      byte[] buffer = new byte[1024];
+      int length = 0;
+      while ((length = is.read(buffer)) != -1) {
+        os.write(buffer, 0, length);
+      }
+    } catch (FileNotFoundException e) {
+      e.printStackTrace();
+    } catch (IOException e) {
+      e.printStackTrace();
+    } finally {
+      try {
+        os.close();
+        is.close();
+      } catch (IOException e) {
+        e.printStackTrace();
+      }
+    }
+
+  }
+}

examples/whisper.android.java/app/src/main/java/com/litongjava/whisper/android/java/utils/WaveEncoder.java

@@ -0,0 +1,105 @@
+package com.litongjava.whisper.android.java.utils;
+
+import java.io.ByteArrayOutputStream;
+import java.io.File;
+import java.io.FileInputStream;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.nio.ByteBuffer;
+import java.nio.ByteOrder;
+import java.nio.ShortBuffer;
+
+public class WaveEncoder {
+
+  public static float[] decodeWaveFile(File file) throws IOException {
+    ByteArrayOutputStream baos = new ByteArrayOutputStream();
+    try (FileInputStream fis = new FileInputStream(file)) {
+      byte[] buffer = new byte[1024];
+      int bytesRead;
+      while ((bytesRead = fis.read(buffer)) != -1) {
+        baos.write(buffer, 0, bytesRead);
+      }
+    }
+    ByteBuffer byteBuffer = ByteBuffer.wrap(baos.toByteArray());
+    byteBuffer.order(ByteOrder.LITTLE_ENDIAN);
+
+    int channel = byteBuffer.getShort(22);
+    byteBuffer.position(44);
+
+    ShortBuffer shortBuffer = byteBuffer.asShortBuffer();
+    short[] shortArray = new short[shortBuffer.limit()];
+    shortBuffer.get(shortArray);
+
+    float[] output = new float[shortArray.length / channel];
+
+    for (int index = 0; index < output.length; index++) {
+      if (channel == 1) {
+        output[index] = Math.max(-1f, Math.min(1f, shortArray[index] / 32767.0f));
+      } else {
+        output[index] = Math.max(-1f, Math.min(1f, (shortArray[2 * index] + shortArray[2 * index + 1]) / 32767.0f / 2.0f));
+      }
+    }
+    return output;
+  }
+
+  public static void encodeWaveFile(File file, short[] data) throws IOException {
+    try (FileOutputStream fos = new FileOutputStream(file)) {
+      fos.write(headerBytes(data.length * 2));
+
+      ByteBuffer buffer = ByteBuffer.allocate(data.length * 2);
+      buffer.order(ByteOrder.LITTLE_ENDIAN);
+      buffer.asShortBuffer().put(data);
+
+      byte[] bytes = new byte[buffer.limit()];
+      buffer.get(bytes);
+
+      fos.write(bytes);
+    }
+  }
+
+  private static byte[] headerBytes(int totalLength) {
+    if (totalLength < 44)
+      throw new IllegalArgumentException("Total length must be at least 44 bytes");
+
+    ByteBuffer buffer = ByteBuffer.allocate(44);
+    buffer.order(ByteOrder.LITTLE_ENDIAN);
+
+    buffer.put((byte) 'R');
+    buffer.put((byte) 'I');
+    buffer.put((byte) 'F');
+    buffer.put((byte) 'F');
+
+    buffer.putInt(totalLength - 8);
+
+    buffer.put((byte) 'W');
+    buffer.put((byte) 'A');
+    buffer.put((byte) 'V');
+    buffer.put((byte) 'E');
+
+    buffer.put((byte) 'f');
+    buffer.put((byte) 'm');
+    buffer.put((byte) 't');
+    buffer.put((byte) ' ');
+
+    buffer.putInt(16);
+    buffer.putShort((short) 1);
+    buffer.putShort((short) 1);
+    buffer.putInt(16000);
+    buffer.putInt(32000);
+    buffer.putShort((short) 2);
+    buffer.putShort((short) 16);
+
+    buffer.put((byte) 'd');
+    buffer.put((byte) 'a');
+    buffer.put((byte) 't');
+    buffer.put((byte) 'a');
+
+    buffer.putInt(totalLength - 44);
+    buffer.position(0);
+
+    byte[] bytes = new byte[buffer.limit()];
+    buffer.get(bytes);
+
+    return bytes;
+  }
+}

examples/whisper.android.java/app/src/main/java/com/whispercpp/java/whisper/CpuInfo.java

@@ -0,0 +1,121 @@
+package com.whispercpp.java.whisper;
+
+import android.os.Build;
+import android.util.Log;
+
+import androidx.annotation.RequiresApi;
+
+import java.io.BufferedReader;
+import java.io.FileReader;
+import java.io.IOException;
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+public class CpuInfo {
+  private static final String LOG_TAG = "WhisperCpuConfig";
+
+  private List<String> lines;
+
+  public CpuInfo(List<String> lines) {
+    this.lines = lines;
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  public int getHighPerfCpuCount0() {
+    try {
+      return getHighPerfCpuCountByFrequencies();
+    } catch (Exception e) {
+      Log.d(LOG_TAG, "Couldn't read CPU frequencies", e);
+      return getHighPerfCpuCountByVariant();
+    }
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  private int getHighPerfCpuCountByFrequencies() {
+    List<Integer> frequencies = getCpuValues("processor", line -> {
+        try {
+          return getMaxCpuFrequency(Integer.parseInt(line.trim()));
+        } catch (IOException e) {
+          e.printStackTrace();
+        }
+        return 0;
+      }
+    );
+    Log.d(LOG_TAG, "Binned cpu frequencies (frequency, count): " + binnedValues(frequencies));
+    return countDroppingMin(frequencies);
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  private int getHighPerfCpuCountByVariant() {
+    List<Integer> variants = getCpuValues("CPU variant", line -> Integer.parseInt(line.trim().substring(line.indexOf("0x") + 2), 16));
+    Log.d(LOG_TAG, "Binned cpu variants (variant, count): " + binnedValues(variants));
+    return countKeepingMin(variants);
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  private Map<Integer, Integer> binnedValues(List<Integer> values) {
+    Map<Integer, Integer> countMap = new HashMap<>();
+    for (int value : values) {
+      countMap.put(value, countMap.getOrDefault(value, 0) + 1);
+    }
+    return countMap;
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  private List<Integer> getCpuValues(String property, Mapper mapper) {
+    List<Integer> values = new ArrayList<>();
+    for (String line : lines) {
+      if (line.startsWith(property)) {
+        values.add(mapper.map(line.substring(line.indexOf(':') + 1)));
+      }
+    }
+    values.sort(Integer::compareTo);
+    return values;
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  private int countDroppingMin(List<Integer> values) {
+    int min = values.stream().mapToInt(i -> i).min().orElse(Integer.MAX_VALUE);
+    return (int) values.stream().filter(value -> value > min).count();
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  private int countKeepingMin(List<Integer> values) {
+    int min = values.stream().mapToInt(i -> i).min().orElse(Integer.MAX_VALUE);
+    return (int) values.stream().filter(value -> value.equals(min)).count();
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  public static int getHighPerfCpuCount() {
+    try {
+      return readCpuInfo().getHighPerfCpuCount0();
+    } catch (Exception e) {
+      Log.d(LOG_TAG, "Couldn't read CPU info", e);
+      return Math.max(Runtime.getRuntime().availableProcessors() - 4, 0);
+    }
+  }
+
+  private static CpuInfo readCpuInfo() throws IOException {
+    try (BufferedReader reader = new BufferedReader(new FileReader("/proc/cpuinfo"))) {
+      List<String> lines = new ArrayList<>();
+      String line;
+      while ((line = reader.readLine()) != null) {
+        lines.add(line);
+      }
+      return new CpuInfo(lines);
+    }
+  }
+
+  private static int getMaxCpuFrequency(int cpuIndex) throws IOException {
+    String path = "/sys/devices/system/cpu/cpu" + cpuIndex + "/cpufreq/cpuinfo_max_freq";
+    try (BufferedReader reader = new BufferedReader(new FileReader(path))) {
+      return Integer.parseInt(reader.readLine());
+    }
+  }
+
+  private interface Mapper {
+    int map(String line);
+  }
+}

examples/whisper.android.java/app/src/main/java/com/whispercpp/java/whisper/WhisperContext.java

@@ -0,0 +1,138 @@
+package com.whispercpp.java.whisper;
+
+import android.content.res.AssetManager;
+import android.os.Build;
+import android.util.Log;
+
+import androidx.annotation.RequiresApi;
+
+import com.litongjava.whisper.android.java.bean.WhisperSegment;
+
+import java.io.InputStream;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.concurrent.Callable;
+import java.util.concurrent.ExecutionException;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Executors;
+
+public class WhisperContext {
+
+  private static final String LOG_TAG = "LibWhisper";
+  private long ptr;
+  private final ExecutorService executorService;
+
+  private WhisperContext(long ptr) {
+    this.ptr = ptr;
+    this.executorService = Executors.newSingleThreadExecutor();
+  }
+
+  public String transcribeData(float[] data) throws ExecutionException, InterruptedException {
+    return executorService.submit(new Callable<String>() {
+      @RequiresApi(api = Build.VERSION_CODES.O)
+      @Override
+      public String call() throws Exception {
+        if (ptr == 0L) {
+          throw new IllegalStateException();
+        }
+        int numThreads = WhisperCpuConfig.getPreferredThreadCount();
+        Log.d(LOG_TAG, "Selecting " + numThreads + " threads");
+
+        StringBuilder result = new StringBuilder();
+        synchronized (this) {
+
+          WhisperLib.fullTranscribe(ptr, numThreads, data);
+          int textCount = WhisperLib.getTextSegmentCount(ptr);
+          for (int i = 0; i < textCount; i++) {
+            String sentence = WhisperLib.getTextSegment(ptr, i);
+            result.append(sentence);
+          }
+        }
+        return result.toString();
+      }
+    }).get();
+  }
+
+  public List<WhisperSegment> transcribeDataWithTime(float[] data) throws ExecutionException, InterruptedException {
+    return executorService.submit(new Callable<List<WhisperSegment>>() {
+      @RequiresApi(api = Build.VERSION_CODES.O)
+      @Override
+      public List<WhisperSegment> call() throws Exception {
+        if (ptr == 0L) {
+          throw new IllegalStateException();
+        }
+        int numThreads = WhisperCpuConfig.getPreferredThreadCount();
+        Log.d(LOG_TAG, "Selecting " + numThreads + " threads");
+
+        List<WhisperSegment> segments = new ArrayList<>();
+        synchronized (this) {
+//          StringBuilder result = new StringBuilder();
+          WhisperLib.fullTranscribe(ptr, numThreads, data);
+          int textCount = WhisperLib.getTextSegmentCount(ptr);
+          for (int i = 0; i < textCount; i++) {
+            long start = WhisperLib.getTextSegmentT0(ptr, i);
+            String sentence = WhisperLib.getTextSegment(ptr, i);
+            long end = WhisperLib.getTextSegmentT1(ptr, i);
+//            result.append();
+            segments.add(new WhisperSegment(start, end, sentence));
+
+          }
+//          return result.toString();
+        }
+        return segments;
+      }
+    }).get();
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public String benchMemory(int nthreads) throws ExecutionException, InterruptedException {
+    return executorService.submit(() -> WhisperLib.benchMemcpy(nthreads)).get();
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public String benchGgmlMulMat(int nthreads) throws ExecutionException, InterruptedException {
+    return executorService.submit(() -> WhisperLib.benchGgmlMulMat(nthreads)).get();
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public void release() throws ExecutionException, InterruptedException {
+    executorService.submit(() -> {
+      if (ptr != 0L) {
+        WhisperLib.freeContext(ptr);
+        ptr = 0;
+      }
+    }).get();
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public static WhisperContext createContextFromFile(String filePath) {
+    long ptr = WhisperLib.initContext(filePath);
+    if (ptr == 0L) {
+      throw new RuntimeException("Couldn't create context with path " + filePath);
+    }
+    return new WhisperContext(ptr);
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public static WhisperContext createContextFromInputStream(InputStream stream) {
+    long ptr = WhisperLib.initContextFromInputStream(stream);
+    if (ptr == 0L) {
+      throw new RuntimeException("Couldn't create context from input stream");
+    }
+    return new WhisperContext(ptr);
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public static WhisperContext createContextFromAsset(AssetManager assetManager, String assetPath) {
+    long ptr = WhisperLib.initContextFromAsset(assetManager, assetPath);
+    if (ptr == 0L) {
+      throw new RuntimeException("Couldn't create context from asset " + assetPath);
+    }
+    return new WhisperContext(ptr);
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public static String getSystemInfo() {
+    return WhisperLib.getSystemInfo();
+  }
+}

examples/whisper.android.java/app/src/main/java/com/whispercpp/java/whisper/WhisperCpuConfig.java

@@ -0,0 +1,12 @@
+package com.whispercpp.java.whisper;
+
+import android.os.Build;
+
+import androidx.annotation.RequiresApi;
+
+public class WhisperCpuConfig {
+  @RequiresApi(api = Build.VERSION_CODES.N)
+  public static int getPreferredThreadCount() {
+    return Math.max(CpuInfo.getHighPerfCpuCount(), 2);
+  }
+}

examples/whisper.android.java/app/src/main/java/com/whispercpp/java/whisper/WhisperLib.java

@@ -0,0 +1,75 @@
+package com.whispercpp.java.whisper;
+
+import android.content.res.AssetManager;
+import android.os.Build;
+import android.util.Log;
+
+import androidx.annotation.RequiresApi;
+
+import java.io.InputStream;
+
+@RequiresApi(api = Build.VERSION_CODES.O)
+public class WhisperLib {
+  private static final String LOG_TAG = "LibWhisper";
+
+  static {
+
+    Log.d(LOG_TAG, "Primary ABI: " + Build.SUPPORTED_ABIS[0]);
+    boolean loadVfpv4 = false;
+    boolean loadV8fp16 = false;
+    if (WhisperUtils.isArmEabiV7a()) {
+      String cpuInfo = WhisperUtils.cpuInfo();
+      if (cpuInfo != null) {
+        Log.d(LOG_TAG, "CPU info: " + cpuInfo);
+        if (cpuInfo.contains("vfpv4")) {
+          Log.d(LOG_TAG, "CPU supports vfpv4");
+          loadVfpv4 = true;
+        }
+      }
+    } else if (WhisperUtils.isArmEabiV8a()) {
+      String cpuInfo = WhisperUtils.cpuInfo();
+      if (cpuInfo != null) {
+        Log.d(LOG_TAG, "CPU info: " + cpuInfo);
+        if (cpuInfo.contains("fphp")) {
+          Log.d(LOG_TAG, "CPU supports fp16 arithmetic");
+          loadV8fp16 = true;
+        }
+      }
+    }
+
+    if (loadVfpv4) {
+      Log.d(LOG_TAG, "Loading libwhisper_vfpv4.so");
+      System.loadLibrary("whisper_vfpv4");
+    } else if (loadV8fp16) {
+      Log.d(LOG_TAG, "Loading libwhisper_v8fp16_va.so");
+      System.loadLibrary("whisper_v8fp16_va");
+    } else {
+      Log.d(LOG_TAG, "Loading libwhisper.so");
+      System.loadLibrary("whisper");
+    }
+  }
+
+  public static native long initContextFromInputStream(InputStream inputStream);
+
+  public static native long initContextFromAsset(AssetManager assetManager, String assetPath);
+
+  public static native long initContext(String modelPath);
+
+  public static native void freeContext(long contextPtr);
+
+  public static native void fullTranscribe(long contextPtr, int numThreads, float[] audioData);
+
+  public static native int getTextSegmentCount(long contextPtr);
+
+  public static native String getTextSegment(long contextPtr, int index);
+
+  public static native long getTextSegmentT0(long contextPtr, int index);
+
+  public static native long getTextSegmentT1(long contextPtr, int index);
+
+  public static native String getSystemInfo();
+
+  public static native String benchMemcpy(int nthread);
+
+  public static native String benchGgmlMulMat(int nthread);
+}

examples/whisper.android.java/app/src/main/java/com/whispercpp/java/whisper/WhisperUtils.java

@@ -0,0 +1,34 @@
+package com.whispercpp.java.whisper;
+
+import android.os.Build;
+import android.util.Log;
+
+import androidx.annotation.RequiresApi;
+
+import java.io.File;
+import java.nio.file.Path;
+
+public class WhisperUtils {
+  private static final String LOG_TAG = "LibWhisper";
+
+
+  public static boolean isArmEabiV7a() {
+    return Build.SUPPORTED_ABIS[0].equals("armeabi-v7a");
+  }
+
+  public static boolean isArmEabiV8a() {
+    return Build.SUPPORTED_ABIS[0].equals("arm64-v8a");
+  }
+
+  @RequiresApi(api = Build.VERSION_CODES.O)
+  public static String cpuInfo() {
+    try {
+      Path path = new File("/proc/cpuinfo").toPath();
+      return new String(java.nio.file.Files.readAllBytes(path));
+    } catch (Exception e) {
+      Log.w(LOG_TAG, "Couldn't read /proc/cpuinfo", e);
+      return null;
+    }
+
+  }
+}

examples/whisper.android.java/app/src/main/jni/whisper/CMakeLists.txt

@@ -0,0 +1,56 @@
+cmake_minimum_required(VERSION 3.10)
+
+project(whisper.cpp)
+
+set(CMAKE_CXX_STANDARD 11)
+set(WHISPER_LIB_DIR ${CMAKE_SOURCE_DIR}/../../../../../../../)
+
+set(
+        SOURCE_FILES
+        ${WHISPER_LIB_DIR}/ggml.c
+        ${WHISPER_LIB_DIR}/ggml-alloc.c
+        ${WHISPER_LIB_DIR}/ggml-backend.c
+        ${WHISPER_LIB_DIR}/ggml-quants.c
+        ${WHISPER_LIB_DIR}/whisper.cpp
+        ${CMAKE_SOURCE_DIR}/jni.c
+)
+
+find_library(LOG_LIB log)
+
+function(build_library target_name)
+    add_library(
+        ${target_name}
+        SHARED
+        ${SOURCE_FILES}
+    )
+
+    target_link_libraries(${target_name} ${LOG_LIB} android)
+
+    if (${target_name} STREQUAL "whisper_v8fp16_va")
+        target_compile_options(${target_name} PRIVATE -march=armv8.2-a+fp16)
+    elseif (${target_name} STREQUAL "whisper_vfpv4")
+        target_compile_options(${target_name} PRIVATE -mfpu=neon-vfpv4)
+    endif ()
+
+    if (NOT ${CMAKE_BUILD_TYPE} STREQUAL "Debug")
+
+        target_compile_options(${target_name} PRIVATE -O3)
+        target_compile_options(${target_name} PRIVATE -fvisibility=hidden -fvisibility-inlines-hidden)
+        target_compile_options(${target_name} PRIVATE -ffunction-sections -fdata-sections)
+
+        #target_link_options(${target_name} PRIVATE -Wl,--gc-sections)
+        #target_link_options(${target_name} PRIVATE -Wl,--exclude-libs,ALL)
+        #target_link_options(${target_name} PRIVATE -flto)
+
+    endif ()
+endfunction()
+
+build_library("whisper") # Default target
+
+if (${ANDROID_ABI} STREQUAL "arm64-v8a")
+    build_library("whisper_v8fp16_va")
+elseif (${ANDROID_ABI} STREQUAL "armeabi-v7a")
+    build_library("whisper_vfpv4")
+endif ()
+
+include_directories(${WHISPER_LIB_DIR})

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

@@ -0,0 +1,257 @@
+#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"
+#include "ggml.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;
+}
+
+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_whispercpp_java_whisper_WhisperLib_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_whispercpp_java_whisper_WhisperLib_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_whispercpp_java_whisper_WhisperLib_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_from_file(model_path_chars);
+    (*env)->ReleaseStringUTFChars(env, model_path_str, model_path_chars);
+    return (jlong) context;
+}
+
+JNIEXPORT void JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_freeContext(
+        JNIEnv *env, jobject thiz, jlong context_ptr) {
+    UNUSED(env);
+    UNUSED(thiz);
+    struct whisper_context *context = (struct whisper_context *) context_ptr;
+    whisper_free(context);
+}
+
+JNIEXPORT void JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_fullTranscribe(
+        JNIEnv *env, jobject thiz, jlong context_ptr, jint num_threads, jfloatArray audio_data) {
+    UNUSED(thiz);
+    struct whisper_context *context = (struct whisper_context *) context_ptr;
+    jfloat *audio_data_arr = (*env)->GetFloatArrayElements(env, audio_data, NULL);
+    const jsize audio_data_length = (*env)->GetArrayLength(env, audio_data);
+
+    // The below adapted from the Objective-C iOS sample
+    struct whisper_full_params params = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
+    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 = num_threads;
+    params.offset_ms = 0;
+    params.no_context = true;
+    params.single_segment = false;
+
+    whisper_reset_timings(context);
+
+    LOGI("About to run whisper_full");
+    if (whisper_full(context, params, audio_data_arr, audio_data_length) != 0) {
+        LOGI("Failed to run the model");
+    } else {
+        whisper_print_timings(context);
+    }
+    (*env)->ReleaseFloatArrayElements(env, audio_data, audio_data_arr, JNI_ABORT);
+}
+
+JNIEXPORT jint JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_getTextSegmentCount(
+        JNIEnv *env, jobject thiz, jlong context_ptr) {
+    UNUSED(env);
+    UNUSED(thiz);
+    struct whisper_context *context = (struct whisper_context *) context_ptr;
+    return whisper_full_n_segments(context);
+}
+
+
+JNIEXPORT jstring JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_getTextSegment(
+        JNIEnv *env, jobject thiz, jlong context_ptr, jint index) {
+    UNUSED(thiz);
+    struct whisper_context *context = (struct whisper_context *) context_ptr;
+    const char *text = whisper_full_get_segment_text(context, index);
+    jstring string = (*env)->NewStringUTF(env, text);
+    return string;
+}
+
+JNIEXPORT jlong JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_getTextSegmentT0(JNIEnv *env, jobject thiz,jlong context_ptr, jint index) {
+    UNUSED(thiz);
+    struct whisper_context *context = (struct whisper_context *) context_ptr;
+    const int64_t t0 = whisper_full_get_segment_t0(context, index);
+    return (jlong)t0;
+}
+
+JNIEXPORT jlong JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_getTextSegmentT1(JNIEnv *env, jobject thiz,jlong context_ptr, jint index) {
+    UNUSED(thiz);
+    struct whisper_context *context = (struct whisper_context *) context_ptr;
+    const int64_t t1 = whisper_full_get_segment_t1(context, index);
+    return (jlong)t1;
+}
+
+JNIEXPORT jstring JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_getSystemInfo(
+        JNIEnv *env, jobject thiz
+) {
+    UNUSED(thiz);
+    const char *sysinfo = whisper_print_system_info();
+    jstring string = (*env)->NewStringUTF(env, sysinfo);
+    return string;
+}
+
+JNIEXPORT jstring JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_benchMemcpy(JNIEnv *env, jobject thiz,
+                                                                      jint n_threads) {
+    UNUSED(thiz);
+    const char *bench_ggml_memcpy = whisper_bench_memcpy_str(n_threads);
+    jstring string = (*env)->NewStringUTF(env, bench_ggml_memcpy);
+}
+
+JNIEXPORT jstring JNICALL
+Java_com_whispercpp_java_whisper_WhisperLib_benchGgmlMulMat(JNIEnv *env, jobject thiz,
+                                                                          jint n_threads) {
+    UNUSED(thiz);
+    const char *bench_ggml_mul_mat = whisper_bench_ggml_mul_mat_str(n_threads);
+    jstring string = (*env)->NewStringUTF(env, bench_ggml_mul_mat);
+}
+