silero-vad

We support both silero-vad v4 and v5.

Download models files

The following table lists the supported onnx model files of silero-vad in sherpa-onnx.

	Model size	Download URL	Comment
silero_vad.onnx exported by k2-fsa	629 KB	Download	It supports only 16kHz and is exported and maintained by k2-fsa
silero_vad.int8.onnx exported by k2-fsa	208 KB	Download	It supports only 16kHz and is exported and maintained by k2-fsa It is 8-bit quantized.
silero_vad v4	1.72 MB	Download	It supports both 16kHz and 8kHz samples
silero_vad v5	2.22 MB	Download	It supports both 16kHz and 8kHz samples

If you are curious about how we export the silero-vad v4 to onnx, you can have a look at

https://github.com/lovemefan/Silero-vad-pytorch/issues/5

We have reverse engineered the PyTorch source code of silero-vad v4. You can use it to export silero-vad to onnx, to executorch, to RKNN, or to torchscript.

Android examples

	Source code	Pre-built APK URL
Pure VAD	Address	Download
VAD + non-streaming ASR	Address	Download
VAD + real-time ASR	Address	Download

For pure VAD, please see https://k2-fsa.github.io/sherpa/onnx/vad/apk.html and https://github.com/k2-fsa/sherpa-onnx/tree/master/android/SherpaOnnxVad

WebAssembly examples

	URL
Huggingface space	Address
ModelScope space	Address

Source code is available at https://github.com/k2-fsa/sherpa-onnx/tree/master/wasm/vad

For WebAssembly with VAD + ASR, please see https://github.com/k2-fsa/sherpa-onnx/tree/master/wasm/vad-asr

C API examples

Filename	Comment
vad-moonshine-c-api.c	silero-vad with moonshine for speech recognition with a very long file
vad-sense-voice-c-api.c	silero-vad with SenseVoice for speech recognition with a very long file
vad-whisper-c-api.c	silero-vad with Whisper for speech recognition with a very long file

For APIs of different programming languages, please see https://github.com/k2-fsa/sherpa-onnx