mirror of
https://github.com/ggerganov/whisper.cpp.git
synced 2024-12-20 05:07:52 +00:00
command : refactor to split command list & general transcription modes (#331)
This makes it easier to understand if you're looking for only one of the capabilities.
This commit is contained in:
Andy Maloney
GitHub
parent
424c410c42
commit
f00509d57c
@ -510,86 +510,23 @@ std::vector<std::string> read_allowed_commands(const std::string & fname) {
|
||||
return allowed_commands;
|
||||
}
|
||||
|
||||
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 * ctx = whisper_init(params.model.c_str());
|
||||
|
||||
// print some info about the processing
|
||||
{
|
||||
fprintf(stderr, "\n");
|
||||
if (!whisper_is_multilingual(ctx)) {
|
||||
if (params.language != "en" || params.translate) {
|
||||
params.language = "en";
|
||||
params.translate = false;
|
||||
fprintf(stderr, "%s: WARNING: model is not multilingual, ignoring language and translation options\n", __func__);
|
||||
}
|
||||
}
|
||||
fprintf(stderr, "%s: processing, %d threads, lang = %s, task = %s, timestamps = %d ...\n",
|
||||
__func__,
|
||||
params.n_threads,
|
||||
params.language.c_str(),
|
||||
params.translate ? "translate" : "transcribe",
|
||||
params.no_timestamps ? 0 : 1);
|
||||
|
||||
fprintf(stderr, "\n");
|
||||
}
|
||||
|
||||
// init audio
|
||||
|
||||
audio_async audio(30*1000);
|
||||
if (!audio.init(params.capture_id, WHISPER_SAMPLE_RATE)) {
|
||||
fprintf(stderr, "%s: audio.init() failed!\n", __func__);
|
||||
return 1;
|
||||
}
|
||||
|
||||
audio.resume();
|
||||
|
||||
// wait for 1 second to avoid any buffered noise
|
||||
std::this_thread::sleep_for(std::chrono::milliseconds(1000));
|
||||
audio.clear();
|
||||
|
||||
int max_len = 0;
|
||||
|
||||
bool is_running = true;
|
||||
bool have_prompt = false;
|
||||
bool ask_prompt = true;
|
||||
|
||||
float prob0 = 0.0f;
|
||||
float prob = 0.0f;
|
||||
|
||||
std::vector<float> pcmf32_cur;
|
||||
std::vector<float> pcmf32_prompt;
|
||||
|
||||
std::vector<std::string> allowed_commands;
|
||||
std::vector<std::vector<whisper_token>> allowed_tokens;
|
||||
|
||||
std::string k_prompt;
|
||||
std::vector<whisper_token> k_tokens;
|
||||
|
||||
if (!params.commands.empty()) {
|
||||
// command-list mode
|
||||
// guide the transcription to match the most likely command from a provided list
|
||||
int process_command_list(struct whisper_context * ctx, audio_async &audio, const whisper_params ¶ms) {
|
||||
fprintf(stderr, "\n");
|
||||
fprintf(stderr, "%s: guided mode\n", __func__);
|
||||
|
||||
allowed_commands = read_allowed_commands(params.commands);
|
||||
std::vector<std::string> allowed_commands = read_allowed_commands(params.commands);
|
||||
|
||||
if (allowed_commands.empty()) {
|
||||
fprintf(stderr, "%s: error: failed to read allowed commands from '%s'\n", __func__, params.commands.c_str());
|
||||
return 2;
|
||||
}
|
||||
|
||||
int max_len = 0;
|
||||
|
||||
std::vector<std::vector<whisper_token>> allowed_tokens;
|
||||
|
||||
for (const auto & cmd : allowed_commands) {
|
||||
whisper_token tokens[1024];
|
||||
allowed_tokens.emplace_back();
|
||||
@ -623,7 +560,7 @@ int main(int argc, char ** argv) {
|
||||
fprintf(stderr, " ]\n");
|
||||
}
|
||||
|
||||
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 += ", ";
|
||||
@ -633,6 +570,7 @@ int main(int argc, char ** argv) {
|
||||
k_prompt += ". selected word: ";
|
||||
|
||||
// tokenize prompt
|
||||
std::vector<whisper_token> k_tokens;
|
||||
{
|
||||
k_tokens.resize(1024);
|
||||
const int n = whisper_tokenize(ctx, k_prompt.c_str(), k_tokens.data(), 1024);
|
||||
@ -655,12 +593,10 @@ int main(int argc, char ** argv) {
|
||||
fprintf(stderr, "%s: listening for a command ...\n", __func__);
|
||||
fprintf(stderr, "\n");
|
||||
|
||||
} else {
|
||||
fprintf(stderr, "\n");
|
||||
fprintf(stderr, "%s: general-purpose mode\n", __func__);
|
||||
bool is_running = true;
|
||||
|
||||
k_prompt = "Ok Whisper, start listening for commands.";
|
||||
}
|
||||
std::vector<float> pcmf32_cur;
|
||||
std::vector<float> pcmf32_prompt;
|
||||
|
||||
// main loop
|
||||
while (is_running) {
|
||||
@ -679,98 +615,13 @@ int main(int argc, char ** argv) {
|
||||
}
|
||||
|
||||
if (!is_running) {
|
||||
break;
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
// delay
|
||||
std::this_thread::sleep_for(std::chrono::milliseconds(100));
|
||||
|
||||
if (allowed_commands.empty()) {
|
||||
// general-purpose mode
|
||||
// freely transcribe the voice into text
|
||||
|
||||
if (ask_prompt) {
|
||||
fprintf(stdout, "\n");
|
||||
fprintf(stdout, "%s: Say the following phrase: '%s%s%s'\n", __func__, "\033[1m", k_prompt.c_str(), "\033[0m");
|
||||
fprintf(stdout, "\n");
|
||||
|
||||
ask_prompt = false;
|
||||
}
|
||||
|
||||
{
|
||||
int64_t t_ms = 0;
|
||||
|
||||
audio.get(2000, pcmf32_cur);
|
||||
|
||||
if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
|
||||
fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
|
||||
|
||||
if (!have_prompt) {
|
||||
// wait for activation phrase
|
||||
audio.get(params.prompt_ms, pcmf32_cur);
|
||||
|
||||
const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob0, t_ms));
|
||||
|
||||
fprintf(stdout, "%s: Heard '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", txt.c_str(), "\033[0m", (int) t_ms);
|
||||
|
||||
const float sim = similarity(txt, k_prompt);
|
||||
|
||||
if (txt.length() < 0.8*k_prompt.length() || txt.length() > 1.2*k_prompt.length() || sim < 0.8f) {
|
||||
fprintf(stdout, "%s: WARNING: prompt not recognized, try again\n", __func__);
|
||||
ask_prompt = true;
|
||||
} else {
|
||||
fprintf(stdout, "\n");
|
||||
fprintf(stdout, "%s: The prompt has been recognized!\n", __func__);
|
||||
fprintf(stdout, "%s: Waiting for voice commands ...\n", __func__);
|
||||
fprintf(stdout, "\n");
|
||||
|
||||
// save the audio for the prompt
|
||||
pcmf32_prompt = pcmf32_cur;
|
||||
have_prompt = true;
|
||||
}
|
||||
} else {
|
||||
// we have heard the activation phrase, now detect the commands
|
||||
audio.get(params.command_ms, pcmf32_cur);
|
||||
|
||||
// prepend the prompt audio
|
||||
pcmf32_cur.insert(pcmf32_cur.begin(), pcmf32_prompt.begin(), pcmf32_prompt.end());
|
||||
|
||||
const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob, t_ms));
|
||||
|
||||
prob = 100.0f*(prob - prob0);
|
||||
|
||||
//fprintf(stdout, "%s: heard '%s'\n", __func__, txt.c_str());
|
||||
|
||||
// find the prompt in the text
|
||||
float best_sim = 0.0f;
|
||||
size_t best_len = 0;
|
||||
for (int n = 0.8*k_prompt.size(); n <= 1.2*k_prompt.size(); ++n) {
|
||||
const auto prompt = txt.substr(0, n);
|
||||
|
||||
const float sim = similarity(prompt, k_prompt);
|
||||
|
||||
//fprintf(stderr, "%s: prompt = '%s', sim = %f\n", __func__, prompt.c_str(), sim);
|
||||
|
||||
if (sim > best_sim) {
|
||||
best_sim = sim;
|
||||
best_len = n;
|
||||
}
|
||||
}
|
||||
|
||||
const std::string command = ::trim(txt.substr(best_len));
|
||||
|
||||
fprintf(stdout, "%s: Command '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", command.c_str(), "\033[0m", (int) t_ms);
|
||||
fprintf(stdout, "\n");
|
||||
}
|
||||
|
||||
audio.clear();
|
||||
}
|
||||
}
|
||||
} else {
|
||||
// command-list mode
|
||||
// guide the transcription to match the most likely command from a provided list
|
||||
|
||||
audio.get(2000, pcmf32_cur);
|
||||
|
||||
if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
|
||||
@ -834,8 +685,8 @@ int main(int argc, char ** argv) {
|
||||
fprintf(stdout, "\n");
|
||||
for (const auto & cmd : probs_id) {
|
||||
fprintf(stdout, "%s: %s%-*s%s = %f | ", __func__, "\033[1m", max_len, allowed_commands[cmd.second].c_str(), "\033[0m", cmd.first);
|
||||
for (int i = 0; i < (int) allowed_tokens[cmd.second].size(); ++i) {
|
||||
fprintf(stdout, "'%4s' %f ", whisper_token_to_str(ctx, allowed_tokens[cmd.second][i]), probs[allowed_tokens[cmd.second][i]]);
|
||||
for (int token : allowed_tokens[cmd.second]) {
|
||||
fprintf(stdout, "'%4s' %f ", whisper_token_to_str(ctx, token), probs[token]);
|
||||
}
|
||||
fprintf(stdout, "\n");
|
||||
}
|
||||
@ -845,9 +696,12 @@ int main(int argc, char ** argv) {
|
||||
{
|
||||
const auto t_end = std::chrono::high_resolution_clock::now();
|
||||
|
||||
const float prob = probs_id[0].first;
|
||||
const int index = probs_id[0].second;
|
||||
|
||||
fprintf(stdout, "\n");
|
||||
fprintf(stdout, "%s: detected command: %s%s%s | p = %f | t = %d ms\n", __func__,
|
||||
"\033[1m", allowed_commands[probs_id[0].second].c_str(), "\033[0m", probs_id[0].first,
|
||||
"\033[1m", allowed_commands[index].c_str(), "\033[0m", prob,
|
||||
(int) std::chrono::duration_cast<std::chrono::milliseconds>(t_end - t_start).count());
|
||||
fprintf(stdout, "\n");
|
||||
}
|
||||
@ -855,6 +709,191 @@ int main(int argc, char ** argv) {
|
||||
audio.clear();
|
||||
}
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
// general-purpose mode
|
||||
// freely transcribe the voice into text
|
||||
int process_general_transcription(struct whisper_context * ctx, audio_async &audio, const whisper_params ¶ms) {
|
||||
bool is_running = true;
|
||||
bool have_prompt = false;
|
||||
bool ask_prompt = true;
|
||||
|
||||
float prob0 = 0.0f;
|
||||
float prob = 0.0f;
|
||||
|
||||
std::vector<float> pcmf32_cur;
|
||||
std::vector<float> pcmf32_prompt;
|
||||
|
||||
const std::string k_prompt = "Ok Whisper, start listening for commands.";
|
||||
|
||||
fprintf(stderr, "\n");
|
||||
fprintf(stderr, "%s: general-purpose mode\n", __func__);
|
||||
|
||||
// main loop
|
||||
while (is_running) {
|
||||
// handle Ctrl + C
|
||||
{
|
||||
SDL_Event event;
|
||||
while (SDL_PollEvent(&event)) {
|
||||
switch (event.type) {
|
||||
case SDL_QUIT:
|
||||
{
|
||||
is_running = false;
|
||||
} break;
|
||||
default:
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (!is_running) {
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
// delay
|
||||
std::this_thread::sleep_for(std::chrono::milliseconds(100));
|
||||
|
||||
if (ask_prompt) {
|
||||
fprintf(stdout, "\n");
|
||||
fprintf(stdout, "%s: Say the following phrase: '%s%s%s'\n", __func__, "\033[1m", k_prompt.c_str(), "\033[0m");
|
||||
fprintf(stdout, "\n");
|
||||
|
||||
ask_prompt = false;
|
||||
}
|
||||
|
||||
{
|
||||
audio.get(2000, pcmf32_cur);
|
||||
|
||||
if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
|
||||
fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
|
||||
|
||||
int64_t t_ms = 0;
|
||||
|
||||
if (!have_prompt) {
|
||||
// wait for activation phrase
|
||||
audio.get(params.prompt_ms, pcmf32_cur);
|
||||
|
||||
const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob0, t_ms));
|
||||
|
||||
fprintf(stdout, "%s: Heard '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", txt.c_str(), "\033[0m", (int) t_ms);
|
||||
|
||||
const float sim = similarity(txt, k_prompt);
|
||||
|
||||
if (txt.length() < 0.8*k_prompt.length() || txt.length() > 1.2*k_prompt.length() || sim < 0.8f) {
|
||||
fprintf(stdout, "%s: WARNING: prompt not recognized, try again\n", __func__);
|
||||
ask_prompt = true;
|
||||
} else {
|
||||
fprintf(stdout, "\n");
|
||||
fprintf(stdout, "%s: The prompt has been recognized!\n", __func__);
|
||||
fprintf(stdout, "%s: Waiting for voice commands ...\n", __func__);
|
||||
fprintf(stdout, "\n");
|
||||
|
||||
// save the audio for the prompt
|
||||
pcmf32_prompt = pcmf32_cur;
|
||||
have_prompt = true;
|
||||
}
|
||||
} else {
|
||||
// we have heard the activation phrase, now detect the commands
|
||||
audio.get(params.command_ms, pcmf32_cur);
|
||||
|
||||
// prepend the prompt audio
|
||||
pcmf32_cur.insert(pcmf32_cur.begin(), pcmf32_prompt.begin(), pcmf32_prompt.end());
|
||||
|
||||
const auto txt = ::trim(::transcribe(ctx, params, pcmf32_cur, prob, t_ms));
|
||||
|
||||
prob = 100.0f*(prob - prob0);
|
||||
|
||||
//fprintf(stdout, "%s: heard '%s'\n", __func__, txt.c_str());
|
||||
|
||||
// find the prompt in the text
|
||||
float best_sim = 0.0f;
|
||||
size_t best_len = 0;
|
||||
for (int n = 0.8*k_prompt.size(); n <= 1.2*k_prompt.size(); ++n) {
|
||||
const auto prompt = txt.substr(0, n);
|
||||
|
||||
const float sim = similarity(prompt, k_prompt);
|
||||
|
||||
//fprintf(stderr, "%s: prompt = '%s', sim = %f\n", __func__, prompt.c_str(), sim);
|
||||
|
||||
if (sim > best_sim) {
|
||||
best_sim = sim;
|
||||
best_len = n;
|
||||
}
|
||||
}
|
||||
|
||||
const std::string command = ::trim(txt.substr(best_len));
|
||||
|
||||
fprintf(stdout, "%s: Command '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", command.c_str(), "\033[0m", (int) t_ms);
|
||||
fprintf(stdout, "\n");
|
||||
}
|
||||
|
||||
audio.clear();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
int main(int argc, char ** argv) {
|
||||
whisper_params params;
|
||||
|
||||
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 * ctx = whisper_init(params.model.c_str());
|
||||
|
||||
// print some info about the processing
|
||||
{
|
||||
fprintf(stderr, "\n");
|
||||
if (!whisper_is_multilingual(ctx)) {
|
||||
if (params.language != "en" || params.translate) {
|
||||
params.language = "en";
|
||||
params.translate = false;
|
||||
fprintf(stderr, "%s: WARNING: model is not multilingual, ignoring language and translation options\n", __func__);
|
||||
}
|
||||
}
|
||||
fprintf(stderr, "%s: processing, %d threads, lang = %s, task = %s, timestamps = %d ...\n",
|
||||
__func__,
|
||||
params.n_threads,
|
||||
params.language.c_str(),
|
||||
params.translate ? "translate" : "transcribe",
|
||||
params.no_timestamps ? 0 : 1);
|
||||
|
||||
fprintf(stderr, "\n");
|
||||
}
|
||||
|
||||
// init audio
|
||||
|
||||
audio_async audio(30*1000);
|
||||
if (!audio.init(params.capture_id, WHISPER_SAMPLE_RATE)) {
|
||||
fprintf(stderr, "%s: audio.init() failed!\n", __func__);
|
||||
return 1;
|
||||
}
|
||||
|
||||
audio.resume();
|
||||
|
||||
// wait for 1 second to avoid any buffered noise
|
||||
std::this_thread::sleep_for(std::chrono::milliseconds(1000));
|
||||
audio.clear();
|
||||
|
||||
int ret_val = 0;
|
||||
|
||||
if (!params.commands.empty()) {
|
||||
ret_val = process_command_list(ctx, audio, params);
|
||||
} else {
|
||||
ret_val = process_general_transcription(ctx, audio, params);
|
||||
}
|
||||
|
||||
audio.pause();
|
||||
@ -862,5 +901,5 @@ int main(int argc, char ** argv) {
|
||||
whisper_print_timings(ctx);
|
||||
whisper_free(ctx);
|
||||
|
||||
return 0;
|
||||
return ret_val;
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user