Voice Activity Detection - RunAnywhere Documentation

RunAnywhere.vad answers “is someone talking right now”. It is what a voice interface uses to decide when a turn has ended.

final result = await RunAnywhere.vad.detect(AudioInput.wav(bytes));

Signatures

Future<VadResult> detect(AudioInput audio, {VadOptions? options})
VadStream openStream(AudioFormatSpec format, {VadOptions? options})
Stream<VadEvent> detectStream(Stream<AudioInput> audio, {VadOptions? options})

openStream is synchronous here and returns the handle directly, unlike Web where it returns a Future. There is no reset().

Streaming from a microphone

final stream = RunAnywhere.vad.openStream(
  AudioFormatSpec(
    encoding: AudioEncoding.pcmF32Le,
    sampleRate: 16000,
    channels: 1,
  ),
);

// per buffer
stream.pushFrame(AudioFrame(samples: bytes, sampleCount: count));

// when done
stream.flush();
stream.finish();
await stream.close();

Streaming from a Stream

await for (final event in RunAnywhere.vad.detectStream(audioStream)) {
  if (event is VadEventSpeechStart) setState(() => _speaking = true);
  if (event is VadEventSpeechEnd) setState(() => _speaking = false);
}

Options

final options = VadOptions(
  activationThreshold: 0.5,
  minSpeechMs: 250,
  minSilenceMs: 500,
  prefixPaddingMs: 300,
);

Field	Default	Meaning
`activationThreshold`	model default	How confident before it counts as speech
`minSpeechMs`	proto default	Ignore bursts shorter than this
`minSilenceMs`	proto default	Silence before a turn is over
`prefixPaddingMs`	proto default	Audio kept from before speech started

minSilenceMs is the one to tune. Too short and the assistant interrupts someone who paused; too long and it feels sluggish. prefixPaddingMs matters because detection lags the first syllable. Without padding, “hello” reaches the transcriber as “ello”.

You may not need this

For a spoken conversation, use the voice session. It runs VAD, transcription, generation, and speech together, with turn-taking and interruption wired up.