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

Limit the maximum segment length for VAD. (#990)

Browse Source
This commit is contained in:
Fangjun Kuang
2024-06-12 10:49:37 +08:00
committed by GitHub
parent aac86847ad
commit 208da78343
4 changed files with 33 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
sherpa-onnx/csrc/silero-vad-model.cc
View File

@@ -190,6 +190,14 @@ class SileroVadModel::Impl {
int32_t MinSpeechDurationSamples() const { return min_speech_samples_; }
void SetMinSilenceDuration(float s) {
min_silence_samples_ = sample_rate_ * s;
}
void SetThreshold(float threshold) {
config_.silero_vad.threshold = threshold;
}
private:
void Init(void *model_data, size_t model_data_length) {
sess_ = std::make_unique<Ort::Session>(env_, model_data, model_data_length,
@@ -306,4 +314,12 @@ int32_t SileroVadModel::MinSpeechDurationSamples() const {
return impl_->MinSpeechDurationSamples();
}
void SileroVadModel::SetMinSilenceDuration(float s) {
impl_->SetMinSilenceDuration(s);
}
void SileroVadModel::SetThreshold(float threshold) {
impl_->SetThreshold(threshold);
}
} // namespace sherpa_onnx

3
sherpa-onnx/csrc/silero-vad-model.h
View File

@@ -42,6 +42,9 @@ class SileroVadModel : public VadModel {
int32_t MinSilenceDurationSamples() const override;
int32_t MinSpeechDurationSamples() const override;
void SetMinSilenceDuration(float s) override;
void SetThreshold(float threshold) override;
private:
class Impl;
std::unique_ptr<Impl> impl_;

2
sherpa-onnx/csrc/vad-model.h
View File

@@ -42,6 +42,8 @@ class VadModel {
virtual int32_t MinSilenceDurationSamples() const = 0;
virtual int32_t MinSpeechDurationSamples() const = 0;
virtual void SetMinSilenceDuration(float s) = 0;
virtual void SetThreshold(float threshold) = 0;
};
} // namespace sherpa_onnx

12
sherpa-onnx/csrc/voice-activity-detector.cc
View File

@@ -29,6 +29,14 @@ class VoiceActivityDetector::Impl {
#endif
void AcceptWaveform(const float *samples, int32_t n) {
if (buffer_.Size() > max_utterance_length_) {
model_->SetMinSilenceDuration(new_min_silence_duration_s_);
model_->SetThreshold(new_threshold_);
} else {
model_->SetMinSilenceDuration(config_.silero_vad.min_silence_duration);
model_->SetThreshold(config_.silero_vad.threshold);
}
int32_t window_size = model_->WindowSize();
// note n is usually window_size and there is no need to use
@@ -114,6 +122,10 @@ class VoiceActivityDetector::Impl {
CircularBuffer buffer_;
std::vector<float> last_;
int max_utterance_length_ = 16000 * 20; // in samples
float new_min_silence_duration_s_ = 0.1;
float new_threshold_ = 1.10;
int32_t start_ = -1;
};