katsu560
3 years ago
committed by
GitHub
commit
c1808cd641
File diff suppressed because one or more lines are too long
@ -0,0 +1,7 @@
|
||||
if (WHISPER_SUPPORT_SDL2)
|
||||
# command
|
||||
set(TARGET command)
|
||||
add_executable(${TARGET} command.cpp)
|
||||
target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS})
|
||||
target_link_libraries(${TARGET} PRIVATE whisper ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
|
||||
endif ()
|
||||
@ -0,0 +1,28 @@
|
||||
# command
|
||||
|
||||
This is a basic Voice Assistant example that accepts voice commands from the microphone.
|
||||
More info is available in [issue #171](https://github.com/ggerganov/whisper.cpp/issues/171).
|
||||
|
||||
```bash
|
||||
# Run with default arguments and small model
|
||||
./command -m ./models/ggml-small.en.bin -t 8
|
||||
|
||||
# On Raspberry Pi, use tiny or base models + "-ac 768" for better performance
|
||||
./command -m ./models/ggml-tiny.en.bin -ac 768 -t 4 -c 0
|
||||
```
|
||||
|
||||
https://user-images.githubusercontent.com/1991296/204038393-2f846eae-c255-4099-a76d-5735c25c49da.mp4
|
||||
|
||||
## Building
|
||||
|
||||
The `command` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
|
||||
|
||||
```bash
|
||||
# Install SDL2 on Linux
|
||||
sudo apt-get install libsdl2-dev
|
||||
|
||||
# Install SDL2 on Mac OS
|
||||
brew install sdl2
|
||||
|
||||
make command
|
||||
```
|
||||
@ -0,0 +1,655 @@
|
||||
// Voice assistant example
|
||||
//
|
||||
// Speak short text commands to the microphone.
|
||||
// This program will detect your voice command and convert them to text.
|
||||
//
|
||||
// ref: https://github.com/ggerganov/whisper.cpp/issues/171
|
||||
//
|
||||
|
||||
#include "whisper.h"
|
||||
|
||||
#include <SDL.h>
|
||||
#include <SDL_audio.h>
|
||||
|
||||
#include <cassert>
|
||||
#include <cstdio>
|
||||
#include <fstream>
|
||||
#include <mutex>
|
||||
#include <regex>
|
||||
#include <string>
|
||||
#include <thread>
|
||||
#include <vector>
|
||||
|
||||
// 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 = 4000;
|
||||
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;
|
||||
|
||||
bool speed_up = false;
|
||||
bool translate = false;
|
||||
bool no_context = true;
|
||||
bool print_special = false;
|
||||
bool print_energy = false;
|
||||
bool no_timestamps = true;
|
||||
|
||||
std::string language = "en";
|
||||
std::string model = "models/ggml-base.en.bin";
|
||||
std::string fname_out = "";
|
||||
};
|
||||
|
||||
void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
|
||||
|
||||
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 == "-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 {
|
||||
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
|
||||
whisper_print_usage(argc, argv, params);
|
||||
exit(0);
|
||||
}
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
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, " -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, "\n");
|
||||
}
|
||||
|
||||
//
|
||||
// SDL Audio capture
|
||||
//
|
||||
|
||||
class audio_async {
|
||||
public:
|
||||
audio_async(int len_ms);
|
||||
~audio_async();
|
||||
|
||||
bool init(int capture_id, int sample_rate);
|
||||
|
||||
// start capturing audio via the provided SDL callback
|
||||
// keep last len_ms seconds of audio in a circular buffer
|
||||
bool resume();
|
||||
bool pause();
|
||||
bool clear();
|
||||
|
||||
// callback to be called by SDL
|
||||
void callback(uint8_t * stream, int len);
|
||||
|
||||
// get audio data from the circular buffer
|
||||
void get(int ms, std::vector<float> & audio);
|
||||
|
||||
private:
|
||||
SDL_AudioDeviceID m_dev_id_in = 0;
|
||||
|
||||
int m_len_ms = 0;
|
||||
int m_sample_rate = 0;
|
||||
|
||||
bool m_running = false;
|
||||
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;
|
||||
};
|
||||
|
||||
audio_async::audio_async(int len_ms) {
|
||||
m_len_ms = len_ms;
|
||||
}
|
||||
|
||||
audio_async::~audio_async() {
|
||||
if (m_dev_id_in) {
|
||||
SDL_CloseAudioDevice(m_dev_id_in);
|
||||
}
|
||||
}
|
||||
|
||||
bool audio_async::init(int capture_id, int sample_rate) {
|
||||
SDL_LogSetPriority(SDL_LOG_CATEGORY_APPLICATION, SDL_LOG_PRIORITY_INFO);
|
||||
|
||||
if (SDL_Init(SDL_INIT_AUDIO) < 0) {
|
||||
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION, "Couldn't initialize SDL: %s\n", SDL_GetError());
|
||||
return false;
|
||||
}
|
||||
|
||||
SDL_SetHintWithPriority(SDL_HINT_AUDIO_RESAMPLING_MODE, "medium", SDL_HINT_OVERRIDE);
|
||||
|
||||
{
|
||||
int nDevices = SDL_GetNumAudioDevices(SDL_TRUE);
|
||||
fprintf(stderr, "%s: found %d capture devices:\n", __func__, nDevices);
|
||||
for (int i = 0; i < nDevices; i++) {
|
||||
fprintf(stderr, "%s: - Capture device #%d: '%s'\n", __func__, i, SDL_GetAudioDeviceName(i, SDL_TRUE));
|
||||
}
|
||||
}
|
||||
|
||||
SDL_AudioSpec capture_spec_requested;
|
||||
SDL_AudioSpec capture_spec_obtained;
|
||||
|
||||
SDL_zero(capture_spec_requested);
|
||||
SDL_zero(capture_spec_obtained);
|
||||
|
||||
capture_spec_requested.freq = sample_rate;
|
||||
capture_spec_requested.format = AUDIO_F32;
|
||||
capture_spec_requested.channels = 1;
|
||||
capture_spec_requested.samples = 1024;
|
||||
capture_spec_requested.callback = [](void * userdata, uint8_t * stream, int len) {
|
||||
audio_async * audio = (audio_async *) userdata;
|
||||
audio->callback(stream, len);
|
||||
};
|
||||
capture_spec_requested.userdata = this;
|
||||
|
||||
if (capture_id >= 0) {
|
||||
fprintf(stderr, "%s: attempt to open capture device %d : '%s' ...\n", __func__, capture_id, SDL_GetAudioDeviceName(capture_id, SDL_TRUE));
|
||||
m_dev_id_in = SDL_OpenAudioDevice(SDL_GetAudioDeviceName(capture_id, SDL_TRUE), SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
|
||||
} else {
|
||||
fprintf(stderr, "%s: attempt to open default capture device ...\n", __func__);
|
||||
m_dev_id_in = SDL_OpenAudioDevice(nullptr, SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
|
||||
}
|
||||
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: couldn't open an audio device for capture: %s!\n", __func__, SDL_GetError());
|
||||
m_dev_id_in = 0;
|
||||
|
||||
return false;
|
||||
} else {
|
||||
fprintf(stderr, "%s: obtained spec for input device (SDL Id = %d):\n", __func__, m_dev_id_in);
|
||||
fprintf(stderr, "%s: - sample rate: %d\n", __func__, capture_spec_obtained.freq);
|
||||
fprintf(stderr, "%s: - format: %d (required: %d)\n", __func__, capture_spec_obtained.format,
|
||||
capture_spec_requested.format);
|
||||
fprintf(stderr, "%s: - channels: %d (required: %d)\n", __func__, capture_spec_obtained.channels,
|
||||
capture_spec_requested.channels);
|
||||
fprintf(stderr, "%s: - samples per frame: %d\n", __func__, capture_spec_obtained.samples);
|
||||
}
|
||||
|
||||
m_sample_rate = capture_spec_obtained.freq;
|
||||
|
||||
m_audio.resize((m_sample_rate*m_len_ms)/1000);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool audio_async::resume() {
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: no audio device to resume!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
if (m_running) {
|
||||
fprintf(stderr, "%s: already running!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
SDL_PauseAudioDevice(m_dev_id_in, 0);
|
||||
|
||||
m_running = true;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool audio_async::pause() {
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: no audio device to pause!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
if (!m_running) {
|
||||
fprintf(stderr, "%s: already paused!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
SDL_PauseAudioDevice(m_dev_id_in, 1);
|
||||
|
||||
m_running = false;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool audio_async::clear() {
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: no audio device to clear!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
if (!m_running) {
|
||||
fprintf(stderr, "%s: not running!\n", __func__);
|
||||
return false;
|
||||
}
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(m_mutex);
|
||||
|
||||
m_audio_pos = 0;
|
||||
m_audio_len = 0;
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
// callback to be called by SDL
|
||||
void audio_async::callback(uint8_t * stream, int len) {
|
||||
if (!m_running) {
|
||||
return;
|
||||
}
|
||||
|
||||
const size_t n_samples = len / sizeof(float);
|
||||
|
||||
m_audio_new.resize(n_samples);
|
||||
memcpy(m_audio_new.data(), stream, n_samples * sizeof(float));
|
||||
|
||||
//fprintf(stderr, "%s: %zu samples, pos %zu, len %zu\n", __func__, n_samples, m_audio_pos, m_audio_len);
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(m_mutex);
|
||||
|
||||
if (m_audio_pos + n_samples > m_audio.size()) {
|
||||
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));
|
||||
|
||||
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
|
||||
m_audio_len = m_audio.size();
|
||||
} else {
|
||||
memcpy(&m_audio[m_audio_pos], stream, n_samples * sizeof(float));
|
||||
|
||||
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
|
||||
m_audio_len = std::min(m_audio_len + n_samples, m_audio.size());
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void audio_async::get(int ms, std::vector<float> & result) {
|
||||
if (!m_dev_id_in) {
|
||||
fprintf(stderr, "%s: no audio device to get audio from!\n", __func__);
|
||||
return;
|
||||
}
|
||||
|
||||
if (!m_running) {
|
||||
fprintf(stderr, "%s: not running!\n", __func__);
|
||||
return;
|
||||
}
|
||||
|
||||
result.clear();
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(m_mutex);
|
||||
|
||||
if (ms <= 0) {
|
||||
ms = m_len_ms;
|
||||
}
|
||||
|
||||
size_t n_samples = (m_sample_rate * ms) / 1000;
|
||||
if (n_samples > m_audio_len) {
|
||||
n_samples = m_audio_len;
|
||||
}
|
||||
|
||||
result.resize(n_samples);
|
||||
|
||||
int s0 = m_audio_pos - n_samples;
|
||||
if (s0 < 0) {
|
||||
s0 += m_audio.size();
|
||||
}
|
||||
|
||||
if (s0 + n_samples > m_audio.size()) {
|
||||
const size_t n0 = m_audio.size() - s0;
|
||||
|
||||
memcpy(result.data(), &m_audio[s0], n0 * sizeof(float));
|
||||
memcpy(&result[n0], &m_audio[0], (n_samples - n0) * sizeof(float));
|
||||
} else {
|
||||
memcpy(result.data(), &m_audio[s0], n_samples * sizeof(float));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
///////////////////////////
|
||||
|
||||
std::string trim(const std::string & s) {
|
||||
std::regex e("^\\s+|\\s+$");
|
||||
return std::regex_replace(s, e, "");
|
||||
}
|
||||
|
||||
void high_pass_filter(std::vector<float> & data, float cutoff, float sample_rate) {
|
||||
const float rc = 1.0f / (2.0f * M_PI * cutoff);
|
||||
const float dt = 1.0f / sample_rate;
|
||||
const float alpha = dt / (rc + dt);
|
||||
|
||||
float y = data[0];
|
||||
|
||||
for (size_t i = 1; i < data.size(); i++) {
|
||||
y = alpha * (y + data[i] - data[i - 1]);
|
||||
data[i] = y;
|
||||
}
|
||||
}
|
||||
|
||||
bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float vad_thold, float freq_thold, bool verbose) {
|
||||
const int n_samples = pcmf32.size();
|
||||
const int n_samples_last = (sample_rate * last_ms) / 1000;
|
||||
|
||||
if (n_samples_last >= n_samples) {
|
||||
// not enough samples - assume no speech
|
||||
return false;
|
||||
}
|
||||
|
||||
if (freq_thold > 0.0f) {
|
||||
high_pass_filter(pcmf32, freq_thold, sample_rate);
|
||||
}
|
||||
|
||||
float energy_all = 0.0f;
|
||||
float energy_last = 0.0f;
|
||||
|
||||
for (size_t i = 0; i < n_samples; i++) {
|
||||
energy_all += fabsf(pcmf32[i]);
|
||||
|
||||
if (i >= n_samples - n_samples_last) {
|
||||
energy_last += fabsf(pcmf32[i]);
|
||||
}
|
||||
}
|
||||
|
||||
energy_all /= n_samples;
|
||||
energy_last /= n_samples_last;
|
||||
|
||||
if (verbose) {
|
||||
fprintf(stderr, "%s: energy_all: %f, energy_last: %f, vad_thold: %f, freq_thold: %f\n", __func__, energy_all, energy_last, vad_thold, freq_thold);
|
||||
}
|
||||
|
||||
if (energy_last > vad_thold*energy_all) {
|
||||
return false;
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
std::string transcribe(whisper_context * ctx, const whisper_params & params, const std::vector<float> & pcmf32, float & prob, int64_t & t_ms) {
|
||||
const auto t_start = std::chrono::high_resolution_clock::now();
|
||||
|
||||
prob = 0.0f;
|
||||
t_ms = 0;
|
||||
|
||||
whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
|
||||
|
||||
wparams.print_progress = false;
|
||||
wparams.print_special = params.print_special;
|
||||
wparams.print_realtime = false;
|
||||
wparams.print_timestamps = !params.no_timestamps;
|
||||
wparams.translate = params.translate;
|
||||
wparams.no_context = true;
|
||||
wparams.single_segment = true;
|
||||
wparams.max_tokens = 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;
|
||||
|
||||
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);
|
||||
for (int i = 0; i < n_segments; ++i) {
|
||||
const char * text = whisper_full_get_segment_text(ctx, i);
|
||||
|
||||
result += text;
|
||||
|
||||
const int n_tokens = whisper_full_n_tokens(ctx, i);
|
||||
for (int j = 0; j < n_tokens; ++j) {
|
||||
const auto token = whisper_full_get_token_data(ctx, i, j);
|
||||
|
||||
prob += token.p;
|
||||
++prob_n;
|
||||
}
|
||||
}
|
||||
|
||||
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();
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
// compute similarity between two strings using Levenshtein distance
|
||||
float similarity(const std::string & s0, const std::string & s1) {
|
||||
const size_t len0 = s0.size() + 1;
|
||||
const size_t len1 = s1.size() + 1;
|
||||
|
||||
std::vector<int> col(len1, 0);
|
||||
std::vector<int> prevCol(len1, 0);
|
||||
|
||||
for (size_t i = 0; i < len1; i++) {
|
||||
prevCol[i] = i;
|
||||
}
|
||||
|
||||
for (size_t i = 0; i < len0; i++) {
|
||||
col[0] = i;
|
||||
for (size_t j = 1; j < len1; j++) {
|
||||
col[j] = std::min(std::min(1 + col[j - 1], 1 + prevCol[j]), prevCol[j - 1] + (s0[i - 1] == s1[j - 1] ? 0 : 1));
|
||||
}
|
||||
col.swap(prevCol);
|
||||
}
|
||||
|
||||
const float dist = prevCol[len1 - 1];
|
||||
|
||||
return 1.0f - (dist / std::max(s0.size(), s1.size()));
|
||||
}
|
||||
|
||||
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();
|
||||
|
||||
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.";
|
||||
|
||||
// main loop
|
||||
while (is_running) {
|
||||
// handle Ctrl + C
|
||||
{
|
||||
SDL_Event event;
|
||||
while (SDL_PollEvent(&event)) {
|
||||
switch (event.type) {
|
||||
case SDL_QUIT:
|
||||
{
|
||||
is_running = false;
|
||||
} break;
|
||||
default:
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (!is_running) {
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
// delay
|
||||
std::this_thread::sleep_for(std::chrono::milliseconds(100));
|
||||
|
||||
if (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) {
|
||||
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 {
|
||||
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();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
audio.pause();
|
||||
|
||||
whisper_print_timings(ctx);
|
||||
whisper_free(ctx);
|
||||
|
||||
return 0;
|
||||
}
|
||||
@ -0,0 +1,182 @@
|
||||
// Common Javascript functions used by the examples
|
||||
|
||||
function convertTypedArray(src, type) {
|
||||
var buffer = new ArrayBuffer(src.byteLength);
|
||||
var baseView = new src.constructor(buffer).set(src);
|
||||
return new type(buffer);
|
||||
}
|
||||
|
||||
var printTextarea = (function() {
|
||||
var element = document.getElementById('output');
|
||||
if (element) element.alue = ''; // clear browser cache
|
||||
return function(text) {
|
||||
if (arguments.length > 1) text = Array.prototype.slice.call(arguments).join(' ');
|
||||
console.log(text);
|
||||
if (element) {
|
||||
element.value += text + "\n";
|
||||
element.scrollTop = element.scrollHeight; // focus on bottom
|
||||
}
|
||||
};
|
||||
})();
|
||||
|
||||
async function clearCache() {
|
||||
if (confirm('Are you sure you want to clear the cache?\nAll the models will be downloaded again.')) {
|
||||
indexedDB.deleteDatabase(dbName);
|
||||
}
|
||||
}
|
||||
|
||||
// fetch a remote file from remote URL using the Fetch API
|
||||
async function fetchRemote(url, cbProgress, cbPrint) {
|
||||
cbPrint('fetchRemote: downloading with fetch()...');
|
||||
|
||||
const response = await fetch(
|
||||
url,
|
||||
{
|
||||
method: 'GET',
|
||||
headers: {
|
||||
'Content-Type': 'application/octet-stream',
|
||||
},
|
||||
}
|
||||
);
|
||||
|
||||
if (!response.ok) {
|
||||
cbPrint('fetchRemote: failed to fetch ' + url);
|
||||
return;
|
||||
}
|
||||
|
||||
const contentLength = response.headers.get('content-length');
|
||||
const total = parseInt(contentLength, 10);
|
||||
const reader = response.body.getReader();
|
||||
|
||||
var chunks = [];
|
||||
var receivedLength = 0;
|
||||
var progressLast = -1;
|
||||
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
|
||||
if (done) {
|
||||
break;
|
||||
}
|
||||
|
||||
chunks.push(value);
|
||||
receivedLength += value.length;
|
||||
|
||||
if (contentLength) {
|
||||
cbProgress(receivedLength/total);
|
||||
|
||||
var progressCur = Math.round((receivedLength / total) * 10);
|
||||
if (progressCur != progressLast) {
|
||||
cbPrint('fetchRemote: fetching ' + 10*progressCur + '% ...');
|
||||
progressLast = progressCur;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
var position = 0;
|
||||
var chunksAll = new Uint8Array(receivedLength);
|
||||
|
||||
for (var chunk of chunks) {
|
||||
chunksAll.set(chunk, position);
|
||||
position += chunk.length;
|
||||
}
|
||||
|
||||
return chunksAll;
|
||||
}
|
||||
|
||||
// load remote data
|
||||
// - check if the data is already in the IndexedDB
|
||||
// - if not, fetch it from the remote URL and store it in the IndexedDB
|
||||
function loadRemote(url, dst, size_mb, cbProgress, cbReady, cbCancel, cbPrint) {
|
||||
// query the storage quota and print it
|
||||
navigator.storage.estimate().then(function (estimate) {
|
||||
cbPrint('loadRemote: storage quota: ' + estimate.quota + ' bytes');
|
||||
cbPrint('loadRemote: storage usage: ' + estimate.usage + ' bytes');
|
||||
});
|
||||
|
||||
// check if the data is already in the IndexedDB
|
||||
var rq = indexedDB.open(dbName, dbVersion);
|
||||
|
||||
rq.onupgradeneeded = function (event) {
|
||||
var db = event.target.result;
|
||||
if (db.version == 1) {
|
||||
var os = db.createObjectStore('models', { autoIncrement: false });
|
||||
cbPrint('loadRemote: created IndexedDB ' + db.name + ' version ' + db.version);
|
||||
} else {
|
||||
// clear the database
|
||||
var os = event.currentTarget.transaction.objectStore('models');
|
||||
os.clear();
|
||||
cbPrint('loadRemote: cleared IndexedDB ' + db.name + ' version ' + db.version);
|
||||
}
|
||||
};
|
||||
|
||||
rq.onsuccess = function (event) {
|
||||
var db = event.target.result;
|
||||
var tx = db.transaction(['models'], 'readonly');
|
||||
var os = tx.objectStore('models');
|
||||
var rq = os.get(url);
|
||||
|
||||
rq.onsuccess = function (event) {
|
||||
if (rq.result) {
|
||||
cbPrint('loadRemote: "' + url + '" is already in the IndexedDB');
|
||||
cbReady(dst, rq.result);
|
||||
} else {
|
||||
// data is not in the IndexedDB
|
||||
cbPrint('loadRemote: "' + url + '" is not in the IndexedDB');
|
||||
|
||||
// alert and ask the user to confirm
|
||||
if (!confirm(
|
||||
'You are about to download ' + size_mb + ' MB of data.\n' +
|
||||
'The model data will be cached in the browser for future use.\n\n' +
|
||||
'Press OK to continue.')) {
|
||||
cbCancel();
|
||||
return;
|
||||
}
|
||||
|
||||
fetchRemote(url, cbProgress, cbPrint).then(function (data) {
|
||||
if (data) {
|
||||
// store the data in the IndexedDB
|
||||
var rq = indexedDB.open(dbName, dbVersion);
|
||||
rq.onsuccess = function (event) {
|
||||
var db = event.target.result;
|
||||
var tx = db.transaction(['models'], 'readwrite');
|
||||
var os = tx.objectStore('models');
|
||||
var rq = os.put(data, url);
|
||||
|
||||
rq.onsuccess = function (event) {
|
||||
cbPrint('loadRemote: "' + url + '" stored in the IndexedDB');
|
||||
cbReady(dst, data);
|
||||
};
|
||||
|
||||
rq.onerror = function (event) {
|
||||
cbPrint('loadRemote: failed to store "' + url + '" in the IndexedDB');
|
||||
cbCancel();
|
||||
};
|
||||
};
|
||||
}
|
||||
});
|
||||
}
|
||||
};
|
||||
|
||||
rq.onerror = function (event) {
|
||||
cbPrint('loadRemote: failed to get data from the IndexedDB');
|
||||
cbCancel();
|
||||
};
|
||||
};
|
||||
|
||||
rq.onerror = function (event) {
|
||||
cbPrint('loadRemote: failed to open IndexedDB');
|
||||
cbCancel();
|
||||
};
|
||||
|
||||
rq.onblocked = function (event) {
|
||||
cbPrint('loadRemote: failed to open IndexedDB: blocked');
|
||||
cbCancel();
|
||||
};
|
||||
|
||||
rq.onabort = function (event) {
|
||||
cbPrint('loadRemote: failed to open IndexedDB: abort');
|
||||
|
||||
};
|
||||
}
|
||||
|
||||
@ -0,0 +1,47 @@
|
||||
#
|
||||
# libstream
|
||||
#
|
||||
|
||||
set(TARGET libstream)
|
||||
|
||||
add_executable(${TARGET}
|
||||
emscripten.cpp
|
||||
)
|
||||
|
||||
target_link_libraries(${TARGET} PRIVATE
|
||||
whisper
|
||||
)
|
||||
|
||||
unset(EXTRA_FLAGS)
|
||||
|
||||
if (WHISPER_WASM_SINGLE_FILE)
|
||||
set(EXTRA_FLAGS "-s SINGLE_FILE=1")
|
||||
message(STATUS "Embedding WASM inside stream.js")
|
||||
|
||||
add_custom_command(
|
||||
TARGET ${TARGET} POST_BUILD
|
||||
COMMAND ${CMAKE_COMMAND} -E copy
|
||||
${CMAKE_BINARY_DIR}/bin/libstream.js
|
||||
${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/stream.wasm/stream.js
|
||||
)
|
||||
endif()
|
||||
|
||||
set_target_properties(${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']\" \
|
||||
${EXTRA_FLAGS} \
|
||||
")
|
||||
|
||||
#
|
||||
# stream.wasm
|
||||
#
|
||||
|
||||
set(TARGET stream.wasm)
|
||||
|
||||
configure_file(${CMAKE_CURRENT_SOURCE_DIR}/index-tmpl.html ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/index.html @ONLY)
|
||||
configure_file(${CMAKE_CURRENT_SOURCE_DIR}/../helpers.js ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/helpers.js @ONLY)
|
||||
@ -0,0 +1,20 @@
|
||||
# stream.wasm
|
||||
|
||||
Real-time transcription in the browser using WebAssembly
|
||||
|
||||
Online demo: https://whisper.ggerganov.com/stream/
|
||||
|
||||
## Build instructions
|
||||
|
||||
```bash
|
||||
# build using Emscripten (v3.1.2)
|
||||
git clone https://github.com/ggerganov/whisper.cpp
|
||||
cd whisper.cpp
|
||||
mkdir build-em && cd build-em
|
||||
emcmake cmake ..
|
||||
make -j
|
||||
|
||||
# copy the produced page to your HTTP path
|
||||
cp bin/stream.wasm/* /path/to/html/
|
||||
cp bin/libstream.worker.js /path/to/html/
|
||||
```
|
||||
@ -0,0 +1,213 @@
|
||||
#include "ggml.h"
|
||||
#include "whisper.h"
|
||||
|
||||
#include <emscripten.h>
|
||||
#include <emscripten/bind.h>
|
||||
|
||||
#include <atomic>
|
||||
#include <cmath>
|
||||
#include <mutex>
|
||||
#include <string>
|
||||
#include <thread>
|
||||
#include <vector>
|
||||
|
||||
constexpr int N_THREAD = 8;
|
||||
|
||||
std::vector<struct whisper_context *> g_contexts(4, nullptr);
|
||||
|
||||
std::mutex g_mutex;
|
||||
std::thread g_worker;
|
||||
|
||||
std::atomic<bool> g_running(false);
|
||||
|
||||
std::string g_status = "";
|
||||
std::string g_status_forced = "";
|
||||
std::string g_transcribed = "";
|
||||
|
||||
std::vector<float> g_pcmf32;
|
||||
|
||||
void stream_set_status(const std::string & status) {
|
||||
std::lock_guard<std::mutex> lock(g_mutex);
|
||||
g_status = status;
|
||||
}
|
||||
|
||||
void stream_main(size_t index) {
|
||||
stream_set_status("loading data ...");
|
||||
|
||||
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.max_tokens = 32;
|
||||
wparams.audio_ctx = 768; // partial encoder context for better performance
|
||||
|
||||
wparams.language = "en";
|
||||
|
||||
printf("stream: using %d threads\n", N_THREAD);
|
||||
|
||||
std::vector<float> pcmf32;
|
||||
|
||||
// whisper context
|
||||
auto & ctx = g_contexts[index];
|
||||
|
||||
// 5 seconds interval
|
||||
const int64_t window_samples = 5*WHISPER_SAMPLE_RATE;
|
||||
|
||||
while (g_running) {
|
||||
stream_set_status("waiting for audio ...");
|
||||
|
||||
{
|
||||
std::unique_lock<std::mutex> lock(g_mutex);
|
||||
|
||||
if (g_pcmf32.size() < 1024) {
|
||||
lock.unlock();
|
||||
|
||||
std::this_thread::sleep_for(std::chrono::milliseconds(10));
|
||||
|
||||
continue;
|
||||
}
|
||||
|
||||
pcmf32 = std::vector<float>(g_pcmf32.end() - std::min((int64_t) g_pcmf32.size(), window_samples), g_pcmf32.end());
|
||||
g_pcmf32.clear();
|
||||
}
|
||||
|
||||
{
|
||||
const auto t_start = std::chrono::high_resolution_clock::now();
|
||||
|
||||
stream_set_status("running whisper ...");
|
||||
|
||||
int ret = whisper_full(ctx, wparams, pcmf32.data(), pcmf32.size());
|
||||
if (ret != 0) {
|
||||
printf("whisper_full() failed: %d\n", ret);
|
||||
break;
|
||||
}
|
||||
|
||||
const auto t_end = std::chrono::high_resolution_clock::now();
|
||||
|
||||
printf("stream: whisper_full() returned %d in %f seconds\n", ret, std::chrono::duration<double>(t_end - t_start).count());
|
||||
}
|
||||
|
||||
{
|
||||
std::string text_heard;
|
||||
|
||||
{
|
||||
const int n_segments = whisper_full_n_segments(ctx);
|
||||
for (int i = n_segments - 1; 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);
|
||||
|
||||
printf("transcribed: %s\n", text);
|
||||
|
||||
text_heard += text;
|
||||
}
|
||||
}
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(g_mutex);
|
||||
g_transcribed = text_heard;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (index < g_contexts.size()) {
|
||||
whisper_free(g_contexts[index]);
|
||||
g_contexts[index] = nullptr;
|
||||
}
|
||||
}
|
||||
|
||||
EMSCRIPTEN_BINDINGS(stream) {
|
||||
emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
|
||||
for (size_t i = 0; i < g_contexts.size(); ++i) {
|
||||
if (g_contexts[i] == nullptr) {
|
||||
g_contexts[i] = whisper_init(path_model.c_str());
|
||||
if (g_contexts[i] != nullptr) {
|
||||
g_running = true;
|
||||
if (g_worker.joinable()) {
|
||||
g_worker.join();
|
||||
}
|
||||
g_worker = std::thread([i]() {
|
||||
stream_main(i);
|
||||
});
|
||||
|
||||
return i + 1;
|
||||
} else {
|
||||
return (size_t) 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return (size_t) 0;
|
||||
}));
|
||||
|
||||
emscripten::function("free", emscripten::optional_override([](size_t index) {
|
||||
if (g_running) {
|
||||
g_running = false;
|
||||
}
|
||||
}));
|
||||
|
||||
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);
|
||||
}
|
||||
|
||||
return 0;
|
||||
}));
|
||||
|
||||
emscripten::function("get_transcribed", emscripten::optional_override([]() {
|
||||
std::string transcribed;
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(g_mutex);
|
||||
transcribed = std::move(g_transcribed);
|
||||
}
|
||||
|
||||
return transcribed;
|
||||
}));
|
||||
|
||||
emscripten::function("get_status", emscripten::optional_override([]() {
|
||||
std::string status;
|
||||
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(g_mutex);
|
||||
status = g_status_forced.empty() ? g_status : g_status_forced;
|
||||
}
|
||||
|
||||
return status;
|
||||
}));
|
||||
|
||||
emscripten::function("set_status", emscripten::optional_override([](const std::string & status) {
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(g_mutex);
|
||||
g_status_forced = status;
|
||||
}
|
||||
}));
|
||||
}
|
||||
@ -0,0 +1,385 @@
|
||||
<!doctype html>
|
||||
<html lang="en-us">
|
||||
<head>
|
||||
<title>stream : Real-time Whisper transcription in WebAssembly</title>
|
||||
|
||||
<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;
|
||||
}
|
||||
</style>
|
||||
</head>
|
||||
<body>
|
||||
<div id="main-container">
|
||||
<b>stream : Real-time Whisper transcription in WebAssembly</b>
|
||||
|
||||
<br><br>
|
||||
|
||||
You can find more about this project on <a href="https://github.com/ggerganov/whisper.cpp/tree/master/examples/stream.wasm">GitHub</a>.
|
||||
|
||||
<br><br>
|
||||
|
||||
<hr>
|
||||
|
||||
Select the model you would like to use, click the "Start" button and start speaking
|
||||
|
||||
<br><br>
|
||||
|
||||
<div id="model-whisper">
|
||||
Whisper model: <span id="model-whisper-status"></span>
|
||||
<button id="fetch-whisper-tiny-en" onclick="loadWhisper('tiny.en')">tiny.en (75 MB)</button>
|
||||
<button id="fetch-whisper-base-en" onclick="loadWhisper('base.en')">base.en (142 MB)</button>
|
||||
<span id="fetch-whisper-progress"></span>
|
||||
|
||||
<!--
|
||||
<input type="file" id="file" name="file" onchange="loadFile(event, 'whisper.bin')" />
|
||||
-->
|
||||
</div>
|
||||
|
||||
<br>
|
||||
|
||||
<div id="input">
|
||||
<button id="start" onclick="onStart()" disabled>Start</button>
|
||||
<button id="stop" onclick="onStop()" disabled>Stop</button>
|
||||
<button id="clear" onclick="clearCache()">Clear Cache</button>
|
||||
</div>
|
||||
|
||||
<br>
|
||||
|
||||
<div id="state">
|
||||
Status: <b><span id="state-status">not started</span></b>
|
||||
|
||||
<pre id="state-transcribed">[The transcribed text will be displayed here]</pre>
|
||||
</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>To use a fast desktop or laptop computer (i.e. not a mobile phone)</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/stream.wasm">Source Code</a> |
|
||||
</span>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
<script type="text/javascript" src="helpers.js"></script>
|
||||
<script type='text/javascript'>
|
||||
const kRestartRecording_s = 15;
|
||||
const kSampleRate = 16000;
|
||||
|
||||
window.AudioContext = window.AudioContext || window.webkitAudioContext;
|
||||
window.OfflineAudioContext = window.OfflineAudioContext || window.webkitOfflineAudioContext;
|
||||
|
||||
// web audio context
|
||||
var context = null;
|
||||
|
||||
// audio data
|
||||
var audio = null;
|
||||
var audio0 = null;
|
||||
|
||||
// the stream instance
|
||||
var instance = null;
|
||||
|
||||
// model name
|
||||
var model_whisper = 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: Initialized successfully!');
|
||||
}
|
||||
};
|
||||
|
||||
//
|
||||
// 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 + '"!';
|
||||
|
||||
if (model_whisper != null) {
|
||||
document.getElementById('start').disabled = false;
|
||||
document.getElementById('stop' ).disabled = true;
|
||||
}
|
||||
}
|
||||
|
||||
function loadWhisper(model) {
|
||||
let urls = {
|
||||
'tiny.en': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
|
||||
'base.en': 'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
|
||||
};
|
||||
|
||||
let sizes = {
|
||||
'tiny.en': 75,
|
||||
'base.en': 142,
|
||||
};
|
||||
|
||||
let url = urls[model];
|
||||
let dst = 'whisper.bin';
|
||||
let size_mb = sizes[model];
|
||||
|
||||
model_whisper = model;
|
||||
|
||||
document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
|
||||
document.getElementById('fetch-whisper-base-en').style.display = 'none';
|
||||
document.getElementById('model-whisper-status').innerHTML = 'loading "' + model + '" ... ';
|
||||
|
||||
cbProgress = function(p) {
|
||||
let el = document.getElementById('fetch-whisper-progress');
|
||||
el.innerHTML = Math.round(100*p) + '%';
|
||||
};
|
||||
|
||||
cbCancel = function() {
|
||||
var el;
|
||||
el = document.getElementById('fetch-whisper-tiny-en'); if (el) el.style.display = 'inline-block';
|
||||
el = document.getElementById('fetch-whisper-base-en'); if (el) el.style.display = 'inline-block';
|
||||
el = document.getElementById('model-whisper-status'); if (el) el.innerHTML = '';
|
||||
};
|
||||
|
||||
loadRemote(url, dst, size_mb, cbProgress, storeFS, cbCancel, printTextarea);
|
||||
}
|
||||
|
||||
//
|
||||
// microphone
|
||||
//
|
||||
|
||||
var mediaRecorder = null;
|
||||
var doRecording = false;
|
||||
var startTime = 0;
|
||||
|
||||
function stopRecording() {
|
||||
Module.set_status("paused");
|
||||
doRecording = false;
|
||||
audio0 = null;
|
||||
audio = null;
|
||||
context = null;
|
||||
}
|
||||
|
||||
function startRecording() {
|
||||
if (!context) {
|
||||
context = new AudioContext({
|
||||
sampleRate: 16000,
|
||||
channelCount: 1,
|
||||
echoCancellation: false,
|
||||
autoGainControl: true,
|
||||
noiseSuppression: true,
|
||||
});
|
||||
}
|
||||
|
||||
Module.set_status("");
|
||||
|
||||
document.getElementById('start').disabled = true;
|
||||
document.getElementById('stop').disabled = false;
|
||||
|
||||
doRecording = 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);
|
||||
|
||||
if (!context) {
|
||||
return;
|
||||
}
|
||||
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) {
|
||||
audio = renderedBuffer.getChannelData(0);
|
||||
|
||||
//printTextarea('js: audio recorded, size: ' + audio.length + ', old size: ' + (audio0 == null ? 0 : audio0.length));
|
||||
|
||||
var audioAll = new Float32Array(audio0 == null ? audio.length : audio0.length + audio.length);
|
||||
if (audio0 != null) {
|
||||
audioAll.set(audio0, 0);
|
||||
}
|
||||
audioAll.set(audio, audio0 == null ? 0 : audio0.length);
|
||||
|
||||
if (instance) {
|
||||
Module.set_audio(instance, audioAll);
|
||||
}
|
||||
});
|
||||
}, function(e) {
|
||||
audio = null;
|
||||
});
|
||||
}
|
||||
|
||||
reader.readAsArrayBuffer(blob);
|
||||
};
|
||||
|
||||
mediaRecorder.onstop = function(e) {
|
||||
if (doRecording) {
|
||||
setTimeout(function() {
|
||||
startRecording();
|
||||
});
|
||||
}
|
||||
};
|
||||
|
||||
mediaRecorder.start(5000);
|
||||
})
|
||||
.catch(function(err) {
|
||||
printTextarea('js: error getting audio stream: ' + err);
|
||||
});
|
||||
|
||||
var interval = setInterval(function() {
|
||||
if (!doRecording) {
|
||||
clearInterval(interval);
|
||||
mediaRecorder.stop();
|
||||
stream.getTracks().forEach(function(track) {
|
||||
track.stop();
|
||||
});
|
||||
|
||||
document.getElementById('start').disabled = false;
|
||||
document.getElementById('stop').disabled = true;
|
||||
|
||||
mediaRecorder = null;
|
||||
}
|
||||
|
||||
// if audio length is more than kRestartRecording_s seconds, restart recording
|
||||
if (audio != null && audio.length > kSampleRate*kRestartRecording_s) {
|
||||
if (doRecording) {
|
||||
//printTextarea('js: restarting recording');
|
||||
|
||||
clearInterval(interval);
|
||||
audio0 = audio;
|
||||
audio = null;
|
||||
mediaRecorder.stop();
|
||||
stream.getTracks().forEach(function(track) {
|
||||
track.stop();
|
||||
});
|
||||
}
|
||||
}
|
||||
}, 250);
|
||||
}
|
||||
|
||||
//
|
||||
// main
|
||||
//
|
||||
|
||||
var nLines = 0;
|
||||
var intervalUpdate = null;
|
||||
var transcribedAll = '';
|
||||
|
||||
function onStart() {
|
||||
if (!instance) {
|
||||
instance = Module.init('whisper.bin');
|
||||
|
||||
if (instance) {
|
||||
printTextarea("js: whisper initialized, instance: " + instance);
|
||||
}
|
||||
}
|
||||
|
||||
if (!instance) {
|
||||
printTextarea("js: failed to initialize whisper");
|
||||
return;
|
||||
}
|
||||
|
||||
startRecording();
|
||||
|
||||
intervalUpdate = setInterval(function() {
|
||||
var transcribed = Module.get_transcribed();
|
||||
|
||||
if (transcribed != null && transcribed.length > 1) {
|
||||
transcribedAll += transcribed + '<br>';
|
||||
nLines++;
|
||||
|
||||
// if more than 10 lines, remove the first line
|
||||
if (nLines > 10) {
|
||||
var i = transcribedAll.indexOf('<br>');
|
||||
if (i > 0) {
|
||||
transcribedAll = transcribedAll.substring(i + 4);
|
||||
nLines--;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
document.getElementById('state-status').innerHTML = Module.get_status();
|
||||
document.getElementById('state-transcribed').innerHTML = transcribedAll;
|
||||
}, 100);
|
||||
}
|
||||
|
||||
function onStop() {
|
||||
stopRecording();
|
||||
}
|
||||
|
||||
</script>
|
||||
<script type="text/javascript" src="stream.js"></script>
|
||||
</body>
|
||||
</html>
|
||||
@ -1,4 +1,5 @@
|
||||
set(TARGET whisper.wasm)
|
||||
|
||||
configure_file(${CMAKE_CURRENT_SOURCE_DIR}/index-tmpl.html ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/index.html @ONLY)
|
||||
configure_file(${CMAKE_CURRENT_SOURCE_DIR}/../helpers.js ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/helpers.js @ONLY)
|
||||
configure_file(${CMAKE_SOURCE_DIR}/bindings/javascript/whisper.js ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/whisper.js COPYONLY)
|
||||
|
||||
Loading…
Reference in new issue