Skip to main content
Voice Activity Detection (VAD) API is planned for a future release.

​
Overview

Voice Activity Detection (VAD) is a technique for detecting the presence or absence of human speech in an audio signal. It’s commonly used to:
  • Reduce computational load by processing only speech segments
  • Improve speech recognition accuracy by filtering out silence and noise
  • Enable push-to-talk and voice-triggered applications
  • Optimize audio streaming and bandwidth

​
Planned Features

The VAD API will provide:
  • Real-time voice detection: Detect speech in live audio streams
  • Batch processing: Analyze audio files for speech segments
  • Configurable sensitivity: Adjust detection thresholds
  • Multiple VAD models: Support for different VAD architectures
  • Integration with STT: Seamless integration with speech recognition

​
Expected Usage (Preview)

import { createVAD, assetModelPath } from 'react-native-sherpa-onnx/vad';
import { createPcmLiveStream } from 'react-native-sherpa-onnx/audio';

// Create VAD instance
const vad = await createVAD({
  modelPath: assetModelPath('models/silero-vad'),
});

// Process live audio
const mic = createPcmLiveStream({ sampleRate: 16000 });

mic.onData(async (samples, sampleRate) => {
  const isSpeech = await vad.detectSpeech(samples, sampleRate);
  
  if (isSpeech) {
    console.log('Speech detected');
    // Process audio with STT
  }
});

await mic.start();

// Later: cleanup
await mic.stop();
await vad.destroy();

​
Model Support

Planned support for popular VAD models:
  • Silero VAD: Lightweight and accurate VAD model
  • WebRTC VAD: Fast, low-latency detection
  • Custom models: Bring your own ONNX VAD models

​
Integration Example

Combining VAD with streaming STT: 
import { createVAD } from 'react-native-sherpa-onnx/vad';
import { createStreamingSTT } from 'react-native-sherpa-onnx/stt';
import { createPcmLiveStream } from 'react-native-sherpa-onnx/audio';

const vad = await createVAD({ /* ... */ });
const stt = await createStreamingSTT({ /* ... */ });
const stream = await stt.createStream();

const mic = createPcmLiveStream({ sampleRate: 16000 });

mic.onData(async (samples, sampleRate) => {
  const isSpeech = await vad.detectSpeech(samples, sampleRate);
  
  if (isSpeech) {
    // Only process speech segments
    const { result, isEndpoint } = await stream.processAudioChunk(
      samples,
      sampleRate
    );
    
    console.log('Transcription:', result.text);
  }
});

​
Availability

This API is not yet implemented. Track progress on the react-native-sherpa-onnx GitHub repository.

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us