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

Support specifying max speech duration for VAD. (#1348)

Browse Source
This commit is contained in:
Fangjun Kuang
2024-09-14 10:57:46 +08:00
committed by GitHub
parent 5d761712db
commit 1423ddb1f0
5 changed files with 70 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
sherpa-onnx/csrc/voice-activity-detector.cc
View File

@@ -18,14 +18,18 @@ class VoiceActivityDetector::Impl {
explicit Impl(const VadModelConfig &config, float buffer_size_in_seconds = 60)
: model_(VadModel::Create(config)),
config_(config),
buffer_(buffer_size_in_seconds * config.sample_rate) {}
buffer_(buffer_size_in_seconds * config.sample_rate) {
Init();
}
#if __ANDROID_API__ >= 9
Impl(AAssetManager *mgr, const VadModelConfig &config,
float buffer_size_in_seconds = 60)
: model_(VadModel::Create(mgr, config)),
config_(config),
buffer_(buffer_size_in_seconds * config.sample_rate) {}
buffer_(buffer_size_in_seconds * config.sample_rate) {
Init();
}
#endif
void AcceptWaveform(const float *samples, int32_t n) {
@@ -145,6 +149,15 @@ class VoiceActivityDetector::Impl {
const VadModelConfig &GetConfig() const { return config_; }
private:
void Init() {
// TODO(fangjun): Currently, we support only one vad model.
// If a new vad model is added, we need to change the place
// where max_speech_duration is placed.
max_utterance_length_ =
config_.sample_rate * config_.silero_vad.max_speech_duration;
}
private:
std::queue<SpeechSegment> segments_;
@@ -153,9 +166,9 @@ class VoiceActivityDetector::Impl {
CircularBuffer buffer_;
std::vector<float> last_;
int max_utterance_length_ = 16000 * 20; // in samples
int max_utterance_length_ = -1; // in samples
float new_min_silence_duration_s_ = 0.1;
float new_threshold_ = 1.10;
float new_threshold_ = 0.90;
int32_t start_ = -1;
};