Class SileroVad

Readonlyversion

detect

• detect(
      samples: Float32Array,
      opts?: SpeechTimestampOptions,
  ): { probs: Float32Array; segments: VadSegment[] }

  Probabilities + segments in one call.

  Parameters

  • samples: Float32Array
  • opts: SpeechTimestampOptions = {}

  Returns { probs: Float32Array; segments: VadSegment[] }

dispose

• dispose(): void

  Returns void

probs

• probs(samples: Float32Array): Float32Array

  Speech probability per 512-sample chunk (the last chunk zero-padded).

  Parameters

  • samples: Float32Array

  Returns Float32Array

probsStreaming

• probsStreaming(
      samples: Float32Array,
      state: VadStreamState | null,
  ): Float32Array

  Streaming variant: state (LSTM h/c + the previous chunk's 64-sample context) carries across calls; pass null to start fresh. Feed whole 512-sample multiples per call (a trailing partial chunk is zero-padded and ends the stream like the reference).

  Parameters

  • samples: Float32Array
  • state: VadStreamState | null

  Returns Float32Array

Staticload

• load(path?: string | null): SileroVad

  Parameters

  • Optionalpath: string | null

  Returns SileroVad

StaticspeechTimestamps

• speechTimestamps(
      probs: Float32Array,
      audioLengthSamples: number,
      opts?: SpeechTimestampOptions,
  ): VadSegment[]

  get_speech_timestamps over the probabilities (sample offsets).

  Parameters

  • probs: Float32Array
  • audioLengthSamples: number
  • opts: SpeechTimestampOptions = {}

  Returns VadSegment[]

StaticstreamState

• streamState(): VadStreamState

  Fresh streaming state.

  Returns VadStreamState