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

Fix for silero vad v5. (#1065)

Browse Source 
The network input is 64 + 512 samples instead of 512 samples for 16kHz.
This commit is contained in:
Fangjun Kuang
2024-06-30 08:57:23 +08:00
committed by GitHub
parent 61c7eb3063
commit 6cb018184e
3 changed files with 18 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
sherpa-onnx/csrc/voice-activity-detector.cc
View File

@@ -44,13 +44,17 @@ class VoiceActivityDetector::Impl {
// an extra buffer here
last_.insert(last_.end(), samples, samples + n);
if (last_.size() < window_size) {
return;
}
// Note: For v4, window_shift == window_size
int32_t k =
(static_cast<int32_t>(last_.size()) - window_size) / window_shift + 1;
const float *p = last_.data();
bool is_speech = false;
for (int32_t i = 0; i != k; ++i, p += window_shift) {
for (int32_t i = 0; i < k; ++i, p += window_shift) {
buffer_.Push(p, window_shift);
// NOTE(fangjun): Please don't use a very large n.
bool this_window_is_speech = model_->IsSpeech(p, window_size);