The WhisperVadContext class provides methods for detecting speech segments in audio files and data using the Silero VAD model.
Properties
Unique context identifier
Whether GPU/Metal acceleration is active
Explanation if GPU is not available (empty string if GPU is active)
Methods
detectSpeech
Detect speech segments in an audio file or base64-encoded WAV data.
detectSpeech(
filePathOrBase64: string | number,
options?: VadOptions
): Promise<VadSegment[]>
Parameters
Audio source to analyze:
- File path:
'/path/to/audio.wav'
- Base64 WAV:
'data:audio/wav;base64,...'
- Asset:
require('./assets/audio.wav')
Note: Remote URLs (http/https) are not supported. Download the file first. Voice activity detection configuration options. See VadOptions for details. Returns
Array of detected speech segmentsShow VadSegment properties
Start time of speech segment in milliseconds
End time of speech segment in milliseconds
Example
// Detect speech in a file
const segments = await vadContext.detectSpeech('/path/to/audio.wav')
console.log(`Found ${segments.length} speech segments`)
segments.forEach((segment, i) => {
const duration = (segment.t1 - segment.t0) / 1000
console.log(`Segment ${i + 1}: ${segment.t0}ms - ${segment.t1}ms (${duration.toFixed(2)}s)`)
})
// Detect speech with custom options
const segments = await vadContext.detectSpeech('/path/to/audio.wav', {
threshold: 0.6,
minSpeechDurationMs: 500,
minSilenceDurationMs: 200,
})
// Detect speech in base64 WAV data
const base64Audio = 'data:audio/wav;base64,UklGRi...'
const segments = await vadContext.detectSpeech(base64Audio)
detectSpeechData
Detect speech segments in raw audio data (base64-encoded PCM or ArrayBuffer).
detectSpeechData(
audioData: string | ArrayBuffer,
options?: VadOptions
): Promise<VadSegment[]>
Parameters
audioData
string | ArrayBuffer
required
Raw audio data:
- Base64-encoded float32 PCM data (mono, 16kHz)
- ArrayBuffer containing 16-bit PCM data (mono, 16kHz)
When using ArrayBuffer, the data is transferred efficiently via JSI bindings. Voice activity detection configuration options. See VadOptions for details. Returns
Array of detected speech segments with t0 (start time) and t1 (end time) in milliseconds
Example
// Detect speech in ArrayBuffer (efficient JSI transfer)
const audioBuffer: ArrayBuffer = getAudioData() // 16-bit PCM, mono, 16kHz
const segments = await vadContext.detectSpeechData(audioBuffer, {
threshold: 0.5,
minSpeechDurationMs: 250,
})
// Detect speech in base64-encoded PCM data
const base64Pcm = 'AAECAw...' // float32 PCM base64
const segments = await vadContext.detectSpeechData(base64Pcm)
release
Release the VAD context and free its memory.
Example
// Always release the context when finished
await vadContext.release()
// Or use try-finally to ensure cleanup
const vadContext = await initWhisperVad({
filePath: '/path/to/silero_vad.bin',
})
try {
const segments = await vadContext.detectSpeech('/path/to/audio.wav')
// Process segments...
} finally {
await vadContext.release()
}
Complete Example
import { initWhisperVad } from 'whisper.rn'
async function detectSpeechInAudio() {
// Initialize VAD context
const vadContext = await initWhisperVad({
filePath: require('./assets/silero_vad.bin'),
useGpu: true,
})
console.log('GPU enabled:', vadContext.gpu)
if (!vadContext.gpu) {
console.log('Reason:', vadContext.reasonNoGPU)
}
try {
// Detect speech segments
const segments = await vadContext.detectSpeech(
'/path/to/audio.wav',
{
threshold: 0.5,
minSpeechDurationMs: 250,
minSilenceDurationMs: 100,
maxSpeechDurationS: 30,
speechPadMs: 30,
}
)
// Process results
console.log(`Detected ${segments.length} speech segments:`)
segments.forEach((segment, i) => {
const start = segment.t0 / 1000
const end = segment.t1 / 1000
const duration = end - start
console.log(` ${i + 1}. ${start.toFixed(2)}s - ${end.toFixed(2)}s (${duration.toFixed(2)}s)`)
})
} finally {
// Clean up
await vadContext.release()
}
}
- Sample rate: 16kHz
- Channels: Mono (1 channel)
- Format: 16-bit PCM (for ArrayBuffer) or float32 PCM (for base64)
- Supported inputs: WAV files, base64-encoded WAV/PCM, ArrayBuffer
Notes
- ArrayBuffer input uses efficient JSI bindings for better performance
- Remote URLs are not supported; download files before processing
- Always release contexts when finished to avoid memory leaks
- Use
releaseAllWhisperVad() to release all VAD contexts at once