silero-realtime-vad

Real-time Voice Activity Detection (VAD) using Silero ONNX models in Node.js.

Silero VAD

Supports both 8KHz & 16KHz sampling rates

Install

npm install silero-realtime-vad

Audio Chunk Requirements

• For 16 kHz sample rate: send 512 samples per chunk
• For 8 kHz sample rate: send 256 samples per chunk

Chunks should be raw Float32 PCM, normalized between -1.0 and 1.0.

Usage

const vad = new SileroVAD({
  sampleRate: 8000,
  minSpeechDuration: 50,
  minSilenceDuration: 500,
  prefixPaddingDuration: 500,
  maxBufferedSpeech: 5000,
  activationThreshold: 0.4,
  context: true,
})
    .on("SPEECH_STARTED", ({ paddingBuffer }) => {
      console.log("SPEECH STARTED");
      console.log(paddingBuffer);
    })
    .on("SPEECH_ENDED", ({ fullBuffer }) => {
      console.log("SPEECH ENDED");
      console.log(fullBuffer);
    });

const audioChunk = new Float32Array(audioData);

const probability = await vad.processAudio(audioChunk);

License

MIT