Skip to main content
The VadOptions interface defines parameters for configuring voice activity detection behavior when using the Silero VAD model.

​
Interface

interface VadOptions {
  threshold?: number
  minSpeechDurationMs?: number
  minSilenceDurationMs?: number
  maxSpeechDurationS?: number
  speechPadMs?: number
  samplesOverlap?: number
}

​
Properties

​
threshold
number
default:"0.5"
Probability threshold to consider audio as speech.
  • Range: 0.0 to 1.0
  • Higher values = more conservative (fewer false positives)
  • Lower values = more sensitive (may detect more speech)
  • Recommended: 0.4 to 0.7
// Very sensitive - may pick up background noise
{ threshold: 0.3 }

// Balanced - good for most use cases
{ threshold: 0.5 }

// Conservative - only clear speech
{ threshold: 0.7 }
​
minSpeechDurationMs
number
default:"250"
Minimum duration (in milliseconds) for a valid speech segment.Segments shorter than this will be filtered out as noise.
// Filter out very short utterances
{ minSpeechDurationMs: 500 }

// Keep even brief speech
{ minSpeechDurationMs: 100 }
​
minSilenceDurationMs
number
default:"100"
Minimum silence duration (in milliseconds) to consider speech as ended.Short pauses shorter than this will not split speech segments.
// Split on brief pauses
{ minSilenceDurationMs: 50 }

// Tolerate longer pauses within speech
{ minSilenceDurationMs: 300 }
​
maxSpeechDurationS
number
default:"30"
Maximum duration (in seconds) of a speech segment before forcing a new segment.Long continuous speech will be split at this duration to avoid oversized segments.
// Split long speech into 15-second chunks
{ maxSpeechDurationS: 15 }

// Allow longer segments
{ maxSpeechDurationS: 60 }
​
speechPadMs
number
default:"30"
Padding (in milliseconds) added before and after detected speech segments.Helps capture the beginning and end of speech that might be near the detection threshold.
// Minimal padding
{ speechPadMs: 10 }

// Extra padding for safety
{ speechPadMs: 100 }
​
samplesOverlap
number
default:"0.1"
Overlap (in seconds) when copying audio samples from speech segments.Used internally for processing audio chunks with continuity.
// Minimal overlap
{ samplesOverlap: 0.05 }

// More overlap for better continuity
{ samplesOverlap: 0.2 }

​
Usage Examples

​
Default Settings

import { initWhisperVad } from 'whisper.rn'

const vadContext = await initWhisperVad({
  filePath: '/path/to/silero_vad.bin',
})

// Use default VAD settings
const segments = await vadContext.detectSpeech('/path/to/audio.wav')

​
Custom Configuration

// Conservative settings - only clear speech
const segments = await vadContext.detectSpeech('/path/to/audio.wav', {
  threshold: 0.7,
  minSpeechDurationMs: 500,
  minSilenceDurationMs: 200,
  maxSpeechDurationS: 20,
  speechPadMs: 50,
})

​
Sensitive Detection

// Sensitive settings - catch all speech
const segments = await vadContext.detectSpeech('/path/to/audio.wav', {
  threshold: 0.3,
  minSpeechDurationMs: 100,
  minSilenceDurationMs: 50,
  speechPadMs: 100,
})

​
Preset Configurations

// Balanced preset (default)
const balanced: VadOptions = {
  threshold: 0.5,
  minSpeechDurationMs: 250,
  minSilenceDurationMs: 100,
  maxSpeechDurationS: 30,
  speechPadMs: 30,
}

// High accuracy preset
const highAccuracy: VadOptions = {
  threshold: 0.6,
  minSpeechDurationMs: 500,
  minSilenceDurationMs: 200,
  maxSpeechDurationS: 25,
  speechPadMs: 50,
}

// High sensitivity preset
const highSensitivity: VadOptions = {
  threshold: 0.35,
  minSpeechDurationMs: 100,
  minSilenceDurationMs: 50,
  maxSpeechDurationS: 30,
  speechPadMs: 100,
}

const segments = await vadContext.detectSpeech('/path/to/audio.wav', balanced)

​
VadSegment Return Type

The detectSpeech and detectSpeechData methods return an array of VadSegment objects: 
interface VadSegment {
  /** Start time in milliseconds */
  t0: number
  /** End time in milliseconds */
  t1: number
}

​
Example

const segments = await vadContext.detectSpeech('/path/to/audio.wav', {
  threshold: 0.5,
  minSpeechDurationMs: 250,
})

segments.forEach((segment) => {
  console.log(`Speech from ${segment.t0}ms to ${segment.t1}ms`)
  console.log(`Duration: ${segment.t1 - segment.t0}ms`)
})

// Output:
// Speech from 1200ms to 3500ms
// Duration: 2300ms
// Speech from 4800ms to 7100ms
// Duration: 2300ms

​
Tuning Guidelines

​
For Noisy Environments

const noisyEnv: VadOptions = {
  threshold: 0.65,           // Higher threshold
  minSpeechDurationMs: 400,  // Longer minimum duration
  minSilenceDurationMs: 150, // More silence required
  speechPadMs: 50,           // Some padding
}

​
For Quiet, Clear Speech

const clearSpeech: VadOptions = {
  threshold: 0.4,            // Lower threshold
  minSpeechDurationMs: 150,  // Shorter minimum
  minSilenceDurationMs: 80,  // Brief pauses acceptable
  speechPadMs: 30,           // Minimal padding
}

​
For Continuous Speech (Lectures, Podcasts)

const continuous: VadOptions = {
  threshold: 0.5,
  minSpeechDurationMs: 300,
  minSilenceDurationMs: 200,  // Tolerate pauses
  maxSpeechDurationS: 60,     // Allow longer segments
  speechPadMs: 40,
}

​
For Command Words (Short Utterances)

const commands: VadOptions = {
  threshold: 0.5,
  minSpeechDurationMs: 100,   // Very short OK
  minSilenceDurationMs: 100,  // Quick detection
  maxSpeechDurationS: 5,      // Short segments
  speechPadMs: 50,            // Extra padding
}

​
Performance Considerations

  • Lower threshold: More segments detected, more processing time
  • Higher minSpeechDurationMs: Fewer segments, faster processing
  • speechPadMs: Adds to segment duration, increases data to process
  • maxSpeechDurationS: Limits segment size, helps memory management

Build docs developers (and LLMs) love

Get started for freeTalk to us