EngineX-Iluvatar/enginex-mr_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Add CXX API for VAD (#2077)

Browse Source
This commit is contained in:
Fangjun Kuang
2025-04-01 14:51:43 +08:00
committed by GitHub
parent 6ef9aeb8d8
commit 0703bc1b86
10 changed files with 455 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
sherpa-onnx/csrc/CMakeLists.txt
View File

@@ -317,11 +317,12 @@ if(SHERPA_ONNX_ENABLE_BINARY)
add_executable(sherpa-onnx-keyword-spotter sherpa-onnx-keyword-spotter.cc)
add_executable(sherpa-onnx-offline sherpa-onnx-offline.cc)
add_executable(sherpa-onnx-offline-audio-tagging sherpa-onnx-offline-audio-tagging.cc)
add_executable(sherpa-onnx-offline-denoiser sherpa-onnx-offline-denoiser.cc)
add_executable(sherpa-onnx-offline-language-identification sherpa-onnx-offline-language-identification.cc)
add_executable(sherpa-onnx-offline-parallel sherpa-onnx-offline-parallel.cc)
add_executable(sherpa-onnx-offline-punctuation sherpa-onnx-offline-punctuation.cc)
add_executable(sherpa-onnx-online-punctuation sherpa-onnx-online-punctuation.cc)
add_executable(sherpa-onnx-offline-denoiser sherpa-onnx-offline-denoiser.cc)
add_executable(sherpa-onnx-vad sherpa-onnx-vad.cc)
if(SHERPA_ONNX_ENABLE_TTS)
add_executable(sherpa-onnx-offline-tts sherpa-onnx-offline-tts.cc)
@@ -336,11 +337,12 @@ if(SHERPA_ONNX_ENABLE_BINARY)
sherpa-onnx-keyword-spotter
sherpa-onnx-offline
sherpa-onnx-offline-audio-tagging
sherpa-onnx-offline-denoiser
sherpa-onnx-offline-language-identification
sherpa-onnx-offline-parallel
sherpa-onnx-offline-punctuation
sherpa-onnx-offline-denoiser
sherpa-onnx-online-punctuation
sherpa-onnx-vad
)
if(SHERPA_ONNX_ENABLE_TTS)
list(APPEND main_exes

50
sherpa-onnx/csrc/sherpa-onnx-vad-alsa.cc
View File

@@ -7,9 +7,9 @@
#include <stdlib.h>
#include <algorithm>
#include <iomanip>
#include "sherpa-onnx/csrc/alsa.h"
#include "sherpa-onnx/csrc/circular-buffer.h"
#include "sherpa-onnx/csrc/voice-activity-detector.h"
#include "sherpa-onnx/csrc/wave-writer.h"
@@ -84,8 +84,6 @@ as the device_name.
exit(-1);
}
int32_t chunk = 0.1 * alsa.GetActualSampleRate();
auto vad = std::make_unique<sherpa_onnx::VoiceActivityDetector>(config);
fprintf(stderr, "Started. Please speak\n");
@@ -95,36 +93,34 @@ as the device_name.
int32_t k = 0;
while (!stop) {
{
const std::vector<float> &samples = alsa.Read(chunk);
const std::vector<float> &samples = alsa.Read(window_size);
vad->AcceptWaveform(samples.data(), samples.size());
vad->AcceptWaveform(samples.data(), samples.size());
if (vad->IsSpeechDetected() && !printed) {
printed = true;
fprintf(stderr, "\nDetected speech!\n");
}
if (!vad->IsSpeechDetected()) {
printed = false;
}
if (vad->IsSpeechDetected() && !printed) {
printed = true;
fprintf(stderr, "\nDetected speech!\n");
}
if (!vad->IsSpeechDetected()) {
printed = false;
}
while (!vad->Empty()) {
const auto &segment = vad->Front();
float duration =
segment.samples.size() / static_cast<float>(sample_rate);
while (!vad->Empty()) {
const auto &segment = vad->Front();
float duration = segment.samples.size() / static_cast<float>(sample_rate);
fprintf(stderr, "Duration: %.3f seconds\n", duration);
fprintf(stderr, "Duration: %.3f seconds\n", duration);
char filename[128];
snprintf(filename, sizeof(filename), "seg-%d-%.3fs.wav", k, duration);
k += 1;
sherpa_onnx::WriteWave(filename, 16000, segment.samples.data(),
segment.samples.size());
fprintf(stderr, "Saved to %s\n", filename);
fprintf(stderr, "----------\n");
std::ostringstream os;
os << "seg-" << k << "-" << std::fixed << std::setprecision(3) << duration
<< "s.wav";
k += 1;
sherpa_onnx::WriteWave(os.str(), 16000, segment.samples.data(),
segment.samples.size());
fprintf(stderr, "Saved to %s\n", os.str().c_str());
fprintf(stderr, "----------\n");
vad->Pop();
}
vad->Pop();
}
}

110
sherpa-onnx/csrc/sherpa-onnx-vad.cc Normal file
View File

@@ -0,0 +1,110 @@
// sherpa-onnx/csrc/sherpa-onnx-vad.cc
//
// Copyright (c) 2025 Xiaomi Corporation
#include <stdio.h>
#include <stdlib.h>
#include <algorithm>
#include <iomanip>
#include "sherpa-onnx/csrc/voice-activity-detector.h"
#include "sherpa-onnx/csrc/wave-reader.h"
#include "sherpa-onnx/csrc/wave-writer.h"
int32_t main(int32_t argc, char *argv[]) {
const char *kUsageMessage = R"usage(
This program shows how to use VAD in sherpa-onnx
to remove silences from a file.
./bin/sherpa-onnx-vad \
--silero-vad-model=/path/to/silero_vad.onnx \
/path/to/input.wav
/path/to/output.wav
Please download silero_vad.onnx from
https://github.com/snakers4/silero-vad/raw/master/src/silero_vad/data/silero_vad.onnx
For instance, use
wget https://github.com/snakers4/silero-vad/raw/master/src/silero_vad/data/silero_vad.onnx
input.wav should be 16kHz.
)usage";
sherpa_onnx::ParseOptions po(kUsageMessage);
sherpa_onnx::VadModelConfig config;
config.Register(&po);
po.Read(argc, argv);
if (po.NumArgs() != 2) {
fprintf(
stderr,
"Please provide only 2 argument2: the input wav and the output wav\n");
po.PrintUsage();
exit(EXIT_FAILURE);
}
fprintf(stderr, "%s\n", config.ToString().c_str());
if (!config.Validate()) {
fprintf(stderr, "Errors in config!\n");
return -1;
}
std::string wav_filename = po.GetArg(1);
int32_t sampling_rate = -1;
bool is_ok = false;
std::vector<float> samples =
sherpa_onnx::ReadWave(wav_filename, &sampling_rate, &is_ok);
if (!is_ok) {
fprintf(stderr, "Failed to read '%s'\n", wav_filename.c_str());
return -1;
}
if (sampling_rate != 16000) {
fprintf(stderr, "Support only 16000Hz. Given: %d\n", sampling_rate);
return -1;
}
auto vad = std::make_unique<sherpa_onnx::VoiceActivityDetector>(config);
int32_t window_size = config.silero_vad.window_size;
int32_t i = 0;
bool is_eof = false;
std::vector<float> samples_without_silence;
while (!is_eof) {
if (i + window_size < samples.size()) {
vad->AcceptWaveform(samples.data() + i, window_size);
i += window_size;
} else {
vad->Flush();
is_eof = true;
}
while (!vad->Empty()) {
const auto &segment = vad->Front();
float start_time = segment.start / static_cast<float>(sampling_rate);
float end_time = start_time + segment.samples.size() /
static_cast<float>(sampling_rate);
fprintf(stderr, "%.3f -- %.3f\n", start_time, end_time);
samples_without_silence.insert(samples_without_silence.end(),
segment.samples.begin(),
segment.samples.end());
vad->Pop();
}
}
sherpa_onnx::WriteWave(po.GetArg(2), sampling_rate,
samples_without_silence.data(),
samples_without_silence.size());
fprintf(stderr, "Saved to %s\n", po.GetArg(2).c_str());
return 0;
}