Active Speaker & Dominance Router

how the router decides

Each RTCEncodedAudioFrame exposes getMetadata().audioLevel on a 0–127 scale where 0 dBov is the loudest. The router reads it in O(1) with no decode, then makes two decisions per frame:

• Suppress? If the level is below the configured silence floor, drop the forward — no point burning egress bandwidth on near-silence.
• Spotlight? Track the peak level over a sliding window. If a participant is loudest for longer than the active hold, promote them to the spotlight tile.

snippet

// inside an RTCRtpScriptTransform worker for the receiving SFU
self.onrtctransform = (event) => {
  const reader = event.transformer.readable.getReader();
  const writer = event.transformer.writable.getWriter();

  let lastSpotlight = null;
  let holdUntil = 0;
  const FLOOR = 95;       // dBov-ish 0=loudest 127=silent
  const HOLD_MS = 800;

  (async () => {
    while (true) {
      const { value: frame, done } = await reader.read();
      if (done) break;

      const meta = frame.getMetadata();
      const level = meta.audioLevel ?? 127;
      const ssrc  = meta.synchronizationSource;
      const now   = performance.now();

      // 1. suppress silence
      if (level > FLOOR) continue;

      // 2. active-speaker switch
      if (level < (lastSpotlight?.level ?? 127) || now > holdUntil) {
        lastSpotlight = { ssrc, level };
        holdUntil = now + HOLD_MS;
        postMessage({ spotlight: ssrc });
      }

      writer.write(frame);
    }
  })();
};

see also

• Feature page
• Encoded Frame Audio Level Meter
• Silence Suppressor
• Spec: RTCEncodedAudioFrameMetadata.audioLevel
• ChromeStatus entry