Browser VAD (Voice Activity Detection)

A simple browser-based Voice Activity Detection library using Silero VAD.

Installation

npm install browser-vad

Usage

import VAD from 'browser-vad';

// Initialize VAD (must be called before using other methods)
await VAD.initialize();
// Optionally, you can provide a custom model path
// await VAD.initialize('./path/to/silero_vad.onnx');

// Start voice detection
await VAD.start();

// Stop voice detection
VAD.stop();

// Toggle between start and stop
const isRecording = await VAD.toggle();

// Listen for state changes (speaking/not speaking)
VAD.onStateChange((state) => {
    console.log('VAD State:', state);
    // state = {
    //     status: 'active' | 'inactive',
    //     probability: number,
    //     ratio: number
    // }
});

// Listen for audio data
VAD.onDataAvailable((audioBlob) => {
    // audioBlob is a WAV file blob containing the recorded audio
    console.log('Audio data received:', audioBlob);
});

API Reference

VAD.initialize(modelPath?: string): Promise<boolean>

Initializes the VAD system. Must be called before using other methods.

• modelPath: Optional path to the Silero VAD model file. If not provided, uses the default model.
• Returns: Promise that resolves to true if initialization was successful.

VAD.start(): Promise<boolean>

Starts voice detection.

• Returns: Promise that resolves to true if successfully started.

VAD.stop(): void

Stops voice detection.

VAD.toggle(): Promise<boolean>

Toggles between start and stop states.

• Returns: Promise that resolves to the new recording state (true if recording, false if stopped).

VAD.onStateChange(callback: function): void

Sets up a callback for voice activity state changes.

• callback: Function that receives state updates with the following structure:

  {
    status: 'active' | 'inactive',  // active when voice detected
    probability: number,            // confidence score
    ratio: number                  // ratio of voice frames
  }

VAD.onDataAvailable(callback: function): void

Sets up a callback for receiving audio data.

• callback: Function that receives a WAV Blob containing the recorded audio.