public_git
/
whisper.cpp
mirror of https://github.com/ggml-org/whisper.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge 6b9dd6b73a into fc674574ca

This commit is contained in:
Ross Morsali 2026-04-20 10:51:05 +00:00 committed by GitHub
parent fc674574ca 6b9dd6b73a
commit 564b4acfa2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
7 changed files with 1084 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
README.md
View File

@ -509,6 +509,14 @@ cmake --build build -j --config Release
https://user-images.githubusercontent.com/1991296/194935793-76afede7-cfa8-48d8-a80f-28ba83be7d09.mp4
If you want to stream audio from another app without SDL or a microphone device, use
[stream-pcm](examples/stream-pcm). It reads raw PCM from stdin or a pipe:
```bash
ffmpeg -i samples/jfk.wav -f s16le -ac 1 -ar 16000 - | \
./build/bin/whisper-stream-pcm -m ./models/ggml-base.en.bin --format s16 --sample-rate 16000 --step 500 --length 5000
```
## Confidence color-coding
Adding the `--print-colors` argument will print the transcribed text using an experimental color coding strategy
@ -842,6 +850,7 @@ Some of the examples are even ported to run in the browser using WebAssembly. Ch
| [whisper-cli](examples/cli) | [whisper.wasm](examples/whisper.wasm) | Tool for translating and transcribing audio using Whisper |
| [whisper-bench](examples/bench) | [bench.wasm](examples/bench.wasm) | Benchmark the performance of Whisper on your machine |
| [whisper-stream](examples/stream) | [stream.wasm](examples/stream.wasm) | Real-time transcription of raw microphone capture |
| [whisper-stream-pcm](examples/stream-pcm) | | Real-time transcription of raw PCM via stdin/pipe (no SDL dependency) |
| [whisper-command](examples/command) | [command.wasm](examples/command.wasm) | Basic voice assistant example for receiving voice commands from the mic |
| [whisper-server](examples/server) | | HTTP transcription server with OAI-like API |
| [whisper-talk-llama](examples/talk-llama) | | Talk with a LLaMA bot |

1
examples/CMakeLists.txt
View File

@ -107,6 +107,7 @@ else()
add_subdirectory(server)
add_subdirectory(quantize)
add_subdirectory(vad-speech-segments)
add_subdirectory(stream-pcm)
if (WHISPER_SDL2)
add_subdirectory(stream)
add_subdirectory(command)

8
examples/stream-pcm/CMakeLists.txt Normal file
View File

@ -0,0 +1,8 @@
set(TARGET whisper-stream-pcm)
add_executable(${TARGET} stream-pcm.cpp)
include(DefaultTargetOptions)
target_link_libraries(${TARGET} PRIVATE common whisper ${CMAKE_THREAD_LIBS_INIT})
install(TARGETS ${TARGET} RUNTIME)

55
examples/stream-pcm/README.md Normal file
View File

@ -0,0 +1,55 @@
# whisper.cpp/examples/stream-pcm
This example performs real-time inference on raw PCM audio streamed via stdin, a pipe, or a file.
It is PCM-first (input is consumed once) and does not require SDL or a microphone device.
## Usage
Stream raw PCM (16 kHz, mono) into the tool (non-VAD):
```bash
./build/bin/whisper-stream-pcm -m ./models/ggml-base.en.bin --format s16 --sample-rate 16000 --step 1000 --length 10000 --keep 500
```
Enable VAD-based segmentation (optional, recommended for speech bursts):
```bash
./build/bin/whisper-stream-pcm -m ./models/ggml-base.en.bin --format s16 --sample-rate 16000 --vad --vad-probe-ms 200 --vad-silence-ms 800 --vad-pre-roll-ms 300 --length 8000
```
You can also read from a named pipe (FIFO):
```bash
mkfifo /tmp/whisper.pcm
./build/bin/whisper-stream-pcm -m ./models/ggml-base.en.bin --input /tmp/whisper.pcm --format s16 --sample-rate 16000 --step 1000 --length 10000 --keep 500
```
Example of piping a WAV file using ffmpeg (optional, `-re` for realtime pacing):
```bash
ffmpeg -re -i samples/jfk.wav -f s16le -ac 1 -ar 16000 - | \
./build/bin/whisper-stream-pcm -m ./models/ggml-base.en.bin --format s16 --sample-rate 16000 --step 1000 --length 10000 --keep 500
```
Windows (PowerShell + `cmd /c`) pipe example:
```powershell
cmd /c "ffmpeg -re -hide_banner -loglevel error -i samples\jfk.wav -f s16le -ac 1 -ar 16000 - | build-cpu\bin\Release\whisper-stream-pcm.exe -m models\ggml-base.en.bin --format s16 --sample-rate 16000 --step 1000 --length 10000 --keep 500"
```
## Notes
- Input must be raw PCM, mono, 16 kHz. The tool does not resample.
- Supported formats: `f32` or `s16` (little-endian).
- Use `--input -` (default) for stdin.
- `--step` must be > 0 unless `--vad` is enabled.
- For VAD, `--vad-probe-ms` should be at least 200 ms; very small probes can fail to trigger.
## Building
`whisper-stream-pcm` does not depend on SDL and builds with the default examples:
```bash
cmake -B build
cmake --build build --config Release
```

901
examples/stream-pcm/stream-pcm.cpp Normal file
View File

@ -0,0 +1,901 @@
// Real-time speech recognition from a raw PCM stream (stdin or pipe)
//
// This mirrors the mic-based whisper-stream example but reads PCM audio
// from stdin or a file/pipe to avoid external audio device dependencies.
//
#include "common.h"
#include "common-whisper.h"
#include "whisper.h"
#include <algorithm>
#include <atomic>
#include <cstdarg>
#include <chrono>
#include <csignal>
#include <cstdio>
#include <cstring>
#include <ctime>
#include <fstream>
#include <mutex>
#include <string>
#include <thread>
#include <vector>
#if defined(_WIN32)
#include <fcntl.h>
#include <io.h>
#endif
// command-line parameters
struct whisper_params {
int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency());
int32_t step_ms = 3000;
int32_t length_ms = 10000;
int32_t keep_ms = 200;
int32_t max_tokens = 32;
int32_t audio_ctx = 0;
int32_t beam_size = -1;
float vad_thold = 0.6f;
float freq_thold = 100.0f;
bool translate = false;
bool no_fallback = false;
bool print_special = false;
bool no_context = true;
bool no_timestamps = false;
bool tinydiarize = false;
bool save_audio = false; // save audio to wav file
bool use_gpu = true;
bool flash_attn = true;
bool debug = false;
bool use_vad = false;
std::string language = "en";
std::string model = "models/ggml-base.en.bin";
std::string fname_out;
std::string input = "-"; // "-" = stdin
std::string format = "f32"; // f32 or s16
int32_t sample_rate = WHISPER_SAMPLE_RATE;
int32_t vad_probe_ms = 200;
int32_t vad_silence_ms = 800;
int32_t vad_pre_roll_ms = 300;
std::string vad_model = ""; // path to silero .bin model
};
void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
enum class pcm_format {
f32,
s16,
};
static 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 == "--step") { params.step_ms = std::stoi(argv[++i]); }
else if ( arg == "--length") { params.length_ms = std::stoi(argv[++i]); }
else if ( arg == "--keep") { params.keep_ms = 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 == "-bs" || arg == "--beam-size") { params.beam_size = 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 == "-tr" || arg == "--translate") { params.translate = true; }
else if (arg == "-nf" || arg == "--no-fallback") { params.no_fallback = true; }
else if (arg == "-ps" || arg == "--print-special") { params.print_special = true; }
else if (arg == "-kc" || arg == "--keep-context") { params.no_context = 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 == "-tdrz" || arg == "--tinydiarize") { params.tinydiarize = true; }
else if (arg == "-sa" || arg == "--save-audio") { params.save_audio = true; }
else if (arg == "-ng" || arg == "--no-gpu") { params.use_gpu = false; }
else if (arg == "-fa" || arg == "--flash-attn") { params.flash_attn = true; }
else if (arg == "-nfa" || arg == "--no-flash-attn") { params.flash_attn = false; }
else if (arg == "-i" || arg == "--input") { params.input = argv[++i]; }
else if (arg == "--format") { params.format = argv[++i]; }
else if (arg == "--sample-rate") { params.sample_rate = std::stoi(argv[++i]); }
else if (arg == "--debug") { params.debug = true; }
else if (arg == "--vad") { params.use_vad = true; }
else if (arg == "--vad-probe-ms") { params.vad_probe_ms = std::stoi(argv[++i]); }
else if (arg == "--vad-silence-ms") { params.vad_silence_ms = std::stoi(argv[++i]); }
else if (arg == "--vad-pre-roll-ms") { params.vad_pre_roll_ms = std::stoi(argv[++i]); }
else if (arg == "-vm" || arg == "--vad-model") { params.vad_model = 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, " --step N [%-7d] audio step size in milliseconds\n", params.step_ms);
fprintf(stderr, " --length N [%-7d] audio length in milliseconds\n", params.length_ms);
fprintf(stderr, " --keep N [%-7d] audio to keep from previous step in ms\n", params.keep_ms);
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, " -bs N, --beam-size N [%-7d] beam size for beam search\n", params.beam_size);
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, " -tr, --translate [%-7s] translate from source language to english\n", params.translate ? "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, " -kc, --keep-context [%-7s] keep context between audio chunks\n", params.no_context ? "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, " -tdrz, --tinydiarize [%-7s] enable tinydiarize (requires a tdrz model)\n", params.tinydiarize ? "true" : "false");
fprintf(stderr, " -sa, --save-audio [%-7s] save the recorded audio to a file\n", params.save_audio ? "true" : "false");
fprintf(stderr, " -ng, --no-gpu [%-7s] disable GPU inference\n", params.use_gpu ? "false" : "true");
fprintf(stderr, " -fa, --flash-attn [%-7s] enable flash attention during inference\n", params.flash_attn ? "true" : "false");
fprintf(stderr, " -nfa, --no-flash-attn [%-7s] disable flash attention during inference\n", params.flash_attn ? "false" : "true");
fprintf(stderr, " -i PATH, --input PATH [%-7s] input path ('-' for stdin)\n", params.input.c_str());
fprintf(stderr, " --format FMT [%-7s] input format: f32 or s16 (little-endian)\n", params.format.c_str());
fprintf(stderr, " --sample-rate N [%-7d] input sample rate (must be 16000)\n", params.sample_rate);
fprintf(stderr, " --debug [%-7s] enable debug logging\n", params.debug ? "true" : "false");
fprintf(stderr, " --vad [%-7s] enable VAD-based segmentation\n", params.use_vad ? "true" : "false");
fprintf(stderr, " --vad-probe-ms N [%-5d] VAD probe chunk size (ms)\n", params.vad_probe_ms);
fprintf(stderr, " --vad-silence-ms N [%-5d] silence duration to end segment (ms)\n", params.vad_silence_ms);
fprintf(stderr, " --vad-pre-roll-ms N[%-5d] audio prepended before VAD trigger (ms)\n", params.vad_pre_roll_ms);
fprintf(stderr, " -vm PATH, --vad-model PATH [%-5s] path to Silero VAD model (enables silero VAD)\n", params.vad_model.c_str());
fprintf(stderr, "\n");
}
class pcm_async {
public:
pcm_async(int len_ms, int sample_rate, pcm_format format)
: m_len_ms(len_ms), m_sample_rate(sample_rate), m_format(format) {
m_running = false;
m_stop = false;
m_eof = false;
}
~pcm_async() {
pause();
}
bool init(const std::string & input_path) {
#if defined(_WIN32)
_setmode(_fileno(stdin), _O_BINARY);
#endif
if (input_path == "-") {
m_in = stdin;
m_owns_input = false;
} else {
m_in = fopen(input_path.c_str(), "rb");
m_owns_input = true;
}
if (!m_in) {
fprintf(stderr, "%s: failed to open input '%s'\n", __func__, input_path.c_str());
return false;
}
m_audio.resize((m_sample_rate*m_len_ms)/1000);
m_audio_pos = 0;
m_audio_len = 0;
return true;
}
bool resume() {
if (!m_in) {
fprintf(stderr, "%s: no input to resume!\n", __func__);
return false;
}
if (m_running) {
fprintf(stderr, "%s: already running!\n", __func__);
return false;
}
m_stop = false;
m_thread = std::thread(&pcm_async::reader_loop, this);
m_running = true;
return true;
}
bool pause() {
if (!m_running) {
return true;
}
m_stop = true;
if (m_owns_input && m_in) {
fclose(m_in);
m_in = nullptr;
}
if (m_thread.joinable()) {
if (m_owns_input) {
m_thread.join();
} else {
m_thread.detach();
}
}
m_running = false;
return true;
}
bool clear() {
std::lock_guard<std::mutex> lock(m_mutex);
m_audio_pos = 0;
m_audio_len = 0;
m_audio_read = 0;
return true;
}
void get(int ms, std::vector<float> & result) {
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 = (int) m_audio_pos - (int) n_samples;
if (s0 < 0) {
s0 += (int) m_audio.size();
}
if (s0 + (int) n_samples > (int) m_audio.size()) {
const size_t n0 = m_audio.size() - (size_t) 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));
}
}
void pop_ms(int ms, std::vector<float> & result) {
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;
}
if (n_samples == 0) {
return;
}
result.resize(n_samples);
size_t s0 = m_audio_read;
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));
}
m_audio_read = (m_audio_read + n_samples) % m_audio.size();
m_audio_len -= n_samples;
}
size_t available_samples() const {
std::lock_guard<std::mutex> lock(m_mutex);
return m_audio_len;
}
bool is_eof() const {
return m_eof.load();
}
private:
void reader_loop() {
const size_t bytes_per_sample = (m_format == pcm_format::f32) ? 4 : 2;
std::vector<uint8_t> buffer(4096);
std::vector<uint8_t> carry;
while (!m_stop) {
size_t n_read = fread(buffer.data(), 1, buffer.size(), m_in);
if (n_read == 0) {
if (feof(m_in)) {
m_eof = true;
}
break;
}
std::vector<uint8_t> data;
data.reserve(carry.size() + n_read);
if (!carry.empty()) {
data.insert(data.end(), carry.begin(), carry.end());
carry.clear();
}
data.insert(data.end(), buffer.begin(), buffer.begin() + n_read);
const size_t total_bytes = data.size();
const size_t n_samples = total_bytes / bytes_per_sample;
const size_t rem = total_bytes % bytes_per_sample;
if (rem > 0) {
carry.insert(carry.end(), data.end() - rem, data.end());
}
if (n_samples == 0) {
continue;
}
std::vector<float> samples(n_samples);
if (m_format == pcm_format::f32) {
for (size_t i = 0; i < n_samples; ++i) {
float v = 0.0f;
memcpy(&v, &data[i * 4], sizeof(float));
samples[i] = v;
}
} else {
for (size_t i = 0; i < n_samples; ++i) {
int16_t v = 0;
memcpy(&v, &data[i * 2], sizeof(int16_t));
samples[i] = v / 32768.0f;
}
}
push_samples(samples.data(), samples.size());
}
}
void push_samples(const float * data, size_t n_samples) {
if (n_samples == 0) {
return;
}
std::lock_guard<std::mutex> lock(m_mutex);
if (n_samples > m_audio.size()) {
data += (n_samples - m_audio.size());
n_samples = m_audio.size();
}
if (n_samples > m_audio.size() - m_audio_len) {
const size_t drop = n_samples - (m_audio.size() - m_audio_len);
m_audio_read = (m_audio_read + drop) % m_audio.size();
m_audio_len -= drop;
}
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], data, n0 * sizeof(float));
memcpy(&m_audio[0], data + n0, (n_samples - n0) * sizeof(float));
} else {
memcpy(&m_audio[m_audio_pos], data, 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());
}
FILE * m_in = nullptr;
bool m_owns_input = false;
int m_len_ms = 0;
int m_sample_rate = 0;
pcm_format m_format = pcm_format::f32;
std::atomic_bool m_running;
std::atomic_bool m_stop;
std::atomic_bool m_eof;
mutable std::mutex m_mutex;
std::vector<float> m_audio;
size_t m_audio_pos = 0;
size_t m_audio_len = 0;
size_t m_audio_read = 0;
std::thread m_thread;
};
static std::atomic_bool g_running(true);
static void signal_handler(int) {
g_running = false;
}
int main(int argc, char ** argv) {
ggml_backend_load_all();
whisper_params params;
if (whisper_params_parse(argc, argv, params) == false) {
return 1;
}
if (params.sample_rate != WHISPER_SAMPLE_RATE) {
fprintf(stderr, "error: only --sample-rate %d is supported (got %d). resample before streaming.\n",
WHISPER_SAMPLE_RATE, params.sample_rate);
return 1;
}
pcm_format input_format = pcm_format::f32;
if (params.format == "f32") {
input_format = pcm_format::f32;
} else if (params.format == "s16") {
input_format = pcm_format::s16;
} else {
fprintf(stderr, "error: unknown --format '%s' (expected f32 or s16)\n", params.format.c_str());
return 1;
}
if (!params.use_vad) {
if (params.step_ms <= 0) {
fprintf(stderr, "error: --step must be > 0 unless --vad is enabled\n");
return 1;
}
params.keep_ms = std::min(params.keep_ms, params.step_ms);
params.length_ms = std::max(params.length_ms, params.step_ms);
} else {
if (params.length_ms <= 0) {
params.length_ms = 5000;
}
params.keep_ms = 0;
}
const bool use_vad = params.use_vad;
const int n_samples_step = use_vad ? 0 : (1e-3*params.step_ms )*WHISPER_SAMPLE_RATE;
const int n_samples_len = (1e-3*params.length_ms)*WHISPER_SAMPLE_RATE;
const int n_samples_keep = (1e-3*params.keep_ms )*WHISPER_SAMPLE_RATE;
const int n_samples_30s = (1e-3*30000.0 )*WHISPER_SAMPLE_RATE;
const int n_new_line = !use_vad ? std::max(1, params.length_ms / params.step_ms - 1) : 1; // number of steps to print new line
params.no_timestamps = !use_vad;
params.no_context |= use_vad;
params.max_tokens = 0;
std::signal(SIGINT, signal_handler);
// init audio
pcm_async audio(params.length_ms, WHISPER_SAMPLE_RATE, input_format);
if (!audio.init(params.input)) {
fprintf(stderr, "%s: audio.init() failed!\n", __func__);
return 1;
}
auto debug_log = [&](const char * fmt, ...) {
if (!params.debug) {
return;
}
va_list args;
va_start(args, fmt);
vfprintf(stderr, fmt, args);
va_end(args);
};
// whisper init
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);
}
struct whisper_context_params cparams = whisper_context_default_params();
cparams.use_gpu = params.use_gpu;
cparams.flash_attn = params.flash_attn;
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 2;
}
// init Silero VAD context (if requested)
struct whisper_vad_context * vad_ctx = nullptr;
if (use_vad && !params.vad_model.empty()) {
struct whisper_vad_context_params vparams = whisper_vad_default_context_params();
vparams.n_threads = params.n_threads;
vparams.use_gpu = false; // Silero VAD is tiny, always runs on CPU
vad_ctx = whisper_vad_init_from_file_with_params(params.vad_model.c_str(), vparams);
if (!vad_ctx) {
fprintf(stderr, "error: failed to init VAD model '%s'\n", params.vad_model.c_str());
whisper_free(ctx);
return 2;
}
}
std::vector<float> pcmf32 (n_samples_30s, 0.0f);
std::vector<float> pcmf32_old;
std::vector<float> pcmf32_new(n_samples_30s, 0.0f);
std::vector<whisper_token> prompt_tokens;
// 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 samples (step = %.1f sec / len = %.1f sec / keep = %.1f sec), %d threads, lang = %s, task = %s, timestamps = %d ...\n",
__func__,
n_samples_step,
float(n_samples_step)/WHISPER_SAMPLE_RATE,
float(n_samples_len )/WHISPER_SAMPLE_RATE,
float(n_samples_keep)/WHISPER_SAMPLE_RATE,
params.n_threads,
params.language.c_str(),
params.translate ? "translate" : "transcribe",
params.no_timestamps ? 0 : 1);
if (!use_vad) {
fprintf(stderr, "%s: n_new_line = %d, no_context = %d\n", __func__, n_new_line, params.no_context);
} else {
fprintf(stderr, "%s: using VAD (%s), will transcribe on speech activity\n", __func__, vad_ctx ? "silero" : "simple");
}
fprintf(stderr, "\n");
}
debug_log("debug: input='%s' format=%s sample_rate=%d step_ms=%d length_ms=%d keep_ms=%d vad=%d probe_ms=%d silence_ms=%d pre_roll_ms=%d last_ms=%d\n",
params.input.c_str(), params.format.c_str(), params.sample_rate, params.step_ms, params.length_ms, params.keep_ms,
params.use_vad ? 1 : 0, params.vad_probe_ms, params.vad_silence_ms, params.vad_pre_roll_ms, std::max(1, std::min(1000, params.vad_probe_ms / 2)));
int n_iter = 0;
bool is_running = true;
std::ofstream fout;
if (params.fname_out.length() > 0) {
fout.open(params.fname_out);
if (!fout.is_open()) {
fprintf(stderr, "%s: failed to open output file '%s'!\n", __func__, params.fname_out.c_str());
return 1;
}
}
wav_writer wavWriter;
// save wav file
if (params.save_audio) {
// Get current date/time for filename
time_t now = time(0);
char buffer[80];
strftime(buffer, sizeof(buffer), "%Y%m%d%H%M%S", localtime(&now));
std::string filename = std::string(buffer) + ".wav";
wavWriter.open(filename, WHISPER_SAMPLE_RATE, 16, 1);
}
audio.resume();
printf("[Start streaming]\n");
fflush(stdout);
int64_t total_samples = 0;
// main audio loop
auto run_inference = [&](const std::vector<float> & audio_buf, int64_t seg_start_sample, int64_t seg_end_sample) -> bool {
if (audio_buf.empty()) {
return true;
}
whisper_full_params wparams = whisper_full_default_params(params.beam_size > 1 ? WHISPER_SAMPLING_BEAM_SEARCH : 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.single_segment = !use_vad;
wparams.max_tokens = params.max_tokens;
wparams.language = params.language.c_str();
wparams.n_threads = params.n_threads;
wparams.beam_search.beam_size = params.beam_size;
wparams.audio_ctx = params.audio_ctx;
wparams.tdrz_enable = params.tinydiarize; // [TDRZ]
// disable temperature fallback
//wparams.temperature_inc = -1.0f;
wparams.temperature_inc = params.no_fallback ? 0.0f : wparams.temperature_inc;
wparams.prompt_tokens = params.no_context ? nullptr : prompt_tokens.data();
wparams.prompt_n_tokens = params.no_context ? 0 : (int) prompt_tokens.size();
if (whisper_full(ctx, wparams, audio_buf.data(), (int) audio_buf.size()) != 0) {
fprintf(stderr, "%s: failed to process audio\n", argv[0]);
return false;
}
// print result;
{
if (!use_vad) {
printf("\33[2K\r");
// print long empty line to clear the previous line
printf("%s", std::string(100, ' ').c_str());
printf("\33[2K\r");
} else {
const int64_t t0_ms = std::max<int64_t>(0, seg_start_sample * 1000 / WHISPER_SAMPLE_RATE);
const int64_t t1_ms = std::max<int64_t>(0, seg_end_sample * 1000 / WHISPER_SAMPLE_RATE);
printf("\n");
printf("### Transcription %d START | t0 = %d ms | t1 = %d ms\n", n_iter, (int) t0_ms, (int) t1_ms);
printf("\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);
if (params.no_timestamps) {
printf("%s", text);
fflush(stdout);
if (params.fname_out.length() > 0) {
fout << text;
}
} else {
const int64_t t0 = whisper_full_get_segment_t0(ctx, i);
const int64_t t1 = whisper_full_get_segment_t1(ctx, i);
std::string output = "[" + to_timestamp(t0, false) + " --> " + to_timestamp(t1, false) + "] " + text;
if (whisper_full_get_segment_speaker_turn_next(ctx, i)) {
output += " [SPEAKER_TURN]";
}
output += "\n";
printf("%s", output.c_str());
fflush(stdout);
if (params.fname_out.length() > 0) {
fout << output;
}
}
}
if (params.fname_out.length() > 0) {
fout << std::endl;
}
if (use_vad) {
printf("\n");
printf("### Transcription %d END\n", n_iter);
}
}
++n_iter;
fflush(stdout);
return true;
};
std::vector<float> pre_roll;
std::vector<float> speech_buf;
int64_t segment_start_sample = 0;
size_t silence_samples = 0;
bool in_speech = false;
const int vad_probe_ms = std::max(1, params.vad_probe_ms);
const int vad_last_ms = std::max(1, std::min(1000, vad_probe_ms / 2));
const size_t vad_pre_roll_samples = (size_t) (WHISPER_SAMPLE_RATE * std::max(0, params.vad_pre_roll_ms) / 1000);
const size_t vad_silence_samples = (size_t) (WHISPER_SAMPLE_RATE * std::max(0, params.vad_silence_ms) / 1000);
const size_t vad_max_segment_samples = (size_t) n_samples_len;
// main audio loop
while (is_running && g_running) {
if (!use_vad) {
const size_t available = audio.available_samples();
if (available < (size_t) n_samples_step) {
if (audio.is_eof()) {
if (available == 0) {
break;
}
} else {
std::this_thread::sleep_for(std::chrono::milliseconds(5));
continue;
}
}
audio.pop_ms(params.step_ms, pcmf32_new);
debug_log("debug: step audio.pop -> %zu samples (available=%zu eof=%d)\n",
pcmf32_new.size(), audio.available_samples(), audio.is_eof() ? 1 : 0);
if (pcmf32_new.empty()) {
if (audio.is_eof()) {
break;
}
continue;
}
total_samples += pcmf32_new.size();
if (params.save_audio && !pcmf32_new.empty()) {
wavWriter.write(pcmf32_new.data(), pcmf32_new.size());
debug_log("debug: save_audio wrote %zu samples (step)\n", pcmf32_new.size());
}
const int n_samples_new = (int) pcmf32_new.size();
const int n_samples_take = std::min((int) pcmf32_old.size(), std::max(0, n_samples_keep + n_samples_len - n_samples_new));
pcmf32.resize(n_samples_new + n_samples_take);
for (int i = 0; i < n_samples_take; i++) {
pcmf32[i] = pcmf32_old[pcmf32_old.size() - n_samples_take + i];
}
memcpy(pcmf32.data() + n_samples_take, pcmf32_new.data(), n_samples_new*sizeof(float));
pcmf32_old = pcmf32;
if (!run_inference(pcmf32, -1, -1)) {
return 6;
}
if (n_iter % n_new_line == 0) {
printf("\n");
// keep part of the audio for next iteration to try to mitigate word boundary issues
if (n_samples_keep > 0 && (int) pcmf32.size() >= n_samples_keep) {
pcmf32_old = std::vector<float>(pcmf32.end() - n_samples_keep, pcmf32.end());
}
// Add tokens of the last full length segment as the prompt
if (!params.no_context) {
prompt_tokens.clear();
const int n_segments = whisper_full_n_segments(ctx);
for (int i = 0; i < n_segments; ++i) {
const int token_count = whisper_full_n_tokens(ctx, i);
for (int j = 0; j < token_count; ++j) {
prompt_tokens.push_back(whisper_full_get_token_id(ctx, i, j));
}
}
}
}
} else {
const size_t available = audio.available_samples();
if (available == 0) {
if (audio.is_eof()) {
if (in_speech && !speech_buf.empty()) {
const int64_t seg_end_sample = segment_start_sample + (int64_t) speech_buf.size();
if (!run_inference(speech_buf, segment_start_sample, seg_end_sample)) {
return 6;
}
speech_buf.clear();
in_speech = false;
}
break;
}
std::this_thread::sleep_for(std::chrono::milliseconds(5));
continue;
}
audio.pop_ms(vad_probe_ms, pcmf32_new);
debug_log("debug: vad probe audio.pop -> %zu samples (available=%zu eof=%d)\n",
pcmf32_new.size(), audio.available_samples(), audio.is_eof() ? 1 : 0);
if (pcmf32_new.empty()) {
continue;
}
total_samples += pcmf32_new.size();
if (params.save_audio && !pcmf32_new.empty()) {
wavWriter.write(pcmf32_new.data(), pcmf32_new.size());
debug_log("debug: save_audio wrote %zu samples (probe)\n", pcmf32_new.size());
}
bool silence;
if (vad_ctx) {
// Silero VAD: run model on probe chunk, average probs
if (whisper_vad_detect_speech(vad_ctx, pcmf32_new.data(), (int) pcmf32_new.size())) {
const int n_probs = whisper_vad_n_probs(vad_ctx);
const float * probs = whisper_vad_probs(vad_ctx);
float avg = 0.0f;
for (int i = 0; i < n_probs; ++i) {
avg += probs[i];
}
avg = (n_probs > 0) ? avg / n_probs : 0.0f;
silence = (avg < params.vad_thold);
debug_log("debug: silero vad avg=%.3f thold=%.2f silence=%d\n", avg, params.vad_thold, silence ? 1 : 0);
} else {
// detect_speech failed — treat as silence to avoid false triggers
silence = true;
debug_log("debug: silero vad detect_speech failed, treating as silence\n");
}
} else {
silence = ::vad_simple(pcmf32_new, WHISPER_SAMPLE_RATE, vad_last_ms, params.vad_thold, params.freq_thold, false);
debug_log("debug: vad silence=%d (last_ms=%d)\n", silence ? 1 : 0, vad_last_ms);
}
if (!in_speech) {
if (!silence) {
in_speech = true;
silence_samples = 0;
speech_buf.clear();
if (!pre_roll.empty()) {
speech_buf.insert(speech_buf.end(), pre_roll.begin(), pre_roll.end());
}
speech_buf.insert(speech_buf.end(), pcmf32_new.begin(), pcmf32_new.end());
segment_start_sample = total_samples - (int64_t) speech_buf.size();
debug_log("debug: vad speech start (segment_start_sample=%lld)\n", (long long) segment_start_sample);
}
} else {
speech_buf.insert(speech_buf.end(), pcmf32_new.begin(), pcmf32_new.end());
if (!silence) {
silence_samples = 0;
} else {
silence_samples += pcmf32_new.size();
}
if (speech_buf.size() >= vad_max_segment_samples || silence_samples >= vad_silence_samples) {
const int64_t seg_end_sample = segment_start_sample + (int64_t) speech_buf.size();
if (!run_inference(speech_buf, segment_start_sample, seg_end_sample)) {
return 6;
}
speech_buf.clear();
in_speech = false;
silence_samples = 0;
debug_log("debug: vad speech end (segment_end_sample=%lld)\n", (long long) seg_end_sample);
}
}
if (vad_pre_roll_samples > 0) {
pre_roll.insert(pre_roll.end(), pcmf32_new.begin(), pcmf32_new.end());
if (pre_roll.size() > vad_pre_roll_samples) {
pre_roll.erase(pre_roll.begin(), pre_roll.end() - vad_pre_roll_samples);
}
}
}
}
audio.pause();
whisper_print_timings(ctx);
whisper_free(ctx);
if (vad_ctx) {
whisper_vad_free(vad_ctx);
}
return 0;
}

10
tests/CMakeLists.txt
View File

@ -110,3 +110,13 @@ target_compile_definitions(${VAD_TEST} PRIVATE
SAMPLE_PATH="${PROJECT_SOURCE_DIR}/samples/jfk.wav")
add_test(NAME ${VAD_TEST} COMMAND ${VAD_TEST})
set_tests_properties(${VAD_TEST} PROPERTIES LABELS "base;en")
if (WHISPER_BUILD_EXAMPLES)
set(TEST_TARGET test-stream-pcm)
add_executable(${TEST_TARGET} ${TEST_TARGET}.cpp)
target_compile_definitions(${TEST_TARGET} PRIVATE
WHISPER_STREAM_PCM_PATH="$<TARGET_FILE:whisper-stream-pcm>"
WHISPER_TEST_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-ggml-tiny.en.bin")
add_test(NAME ${TEST_TARGET} COMMAND ${TEST_TARGET})
set_tests_properties(${TEST_TARGET} PROPERTIES LABELS "tiny;stream")
endif()

100
tests/test-stream-pcm.cpp Normal file
View File

@ -0,0 +1,100 @@
#include <cstdlib>
#include <cstdio>
#include <cerrno>
#include <cstring>
#include <filesystem>
#include <fstream>
#include <string>
#include <vector>
#if defined(_WIN32)
#include <process.h>
#endif
#ifndef WHISPER_STREAM_PCM_PATH
#error "WHISPER_STREAM_PCM_PATH is not defined"
#endif
#ifndef WHISPER_TEST_MODEL_PATH
#error "WHISPER_TEST_MODEL_PATH is not defined"
#endif
static std::filesystem::path temp_pcm_path() {
std::error_code ec;
auto dir = std::filesystem::temp_directory_path(ec);
if (ec) {
return std::filesystem::path("whisper_stream_pcm_test.raw");
}
return dir / "whisper_stream_pcm_test.raw";
}
int main() {
const int sample_rate = 16000;
const int seconds = 2;
const size_t n_samples = (size_t) sample_rate * seconds;
std::vector<float> zeros(n_samples, 0.0f);
auto pcm_path = temp_pcm_path();
pcm_path.make_preferred();
std::ofstream out(pcm_path, std::ios::binary);
if (!out.is_open()) {
fprintf(stderr, "failed to open temp PCM path: %s\n", pcm_path.string().c_str());
return 1;
}
out.write(reinterpret_cast<const char *>(zeros.data()), zeros.size() * sizeof(float));
out.close();
const std::string stream_bin = std::filesystem::path(WHISPER_STREAM_PCM_PATH).make_preferred().string();
const std::string model_path = std::filesystem::path(WHISPER_TEST_MODEL_PATH).make_preferred().string();
std::vector<std::string> args = {
stream_bin,
"-m", model_path,
"--input", pcm_path.string(),
"--format", "f32",
"--sample-rate", "16000",
"--step", "500",
"--length", "2000",
"-t", "1",
"-ng",
};
int rc = 1;
#if defined(_WIN32)
std::vector<const char *> argv;
argv.reserve(args.size() + 1);
for (const auto & arg : args) {
argv.push_back(arg.c_str());
}
argv.push_back(nullptr);
rc = (int) _spawnv(_P_WAIT, stream_bin.c_str(), argv.data());
if (rc == -1) {
fprintf(stderr, "failed to spawn whisper-stream-pcm: %s\n", std::strerror(errno));
rc = 1;
}
#else
std::string cmd;
cmd.reserve(1024);
for (const auto & arg : args) {
if (!cmd.empty()) {
cmd += " ";
}
cmd += "\"";
cmd += arg;
cmd += "\"";
}
rc = std::system(cmd.c_str());
#endif
std::error_code ec;
std::filesystem::remove(pcm_path, ec);
if (rc != 0) {
fprintf(stderr, "whisper-stream-pcm exited with code %d\n", rc);
return 1;
}
return 0;
}