Crate dominant_speaker

Expand description

Pure-Rust dominant speaker identification for WebRTC applications.

This crate implements the three-time-scale subband comparison algorithm described in Volfin & Cohen, “Dominant Speaker Identification for Multipoint Videoconferencing”, IEEE 2012. The implementation follows mediasoup’s C++ ActiveSpeakerObserver for constants and Jitsi’s Java DominantSpeakerIdentification for the overall structure.

Feed it RFC 6464 audio-level observations and it tells you who is talking. No FFI, no WebRTC stack dependency, no unsafe code.

§Quick start

Timestamps are caller-supplied u64 milliseconds — use any epoch you like (e.g. Instant::now().elapsed().as_millis() as u64 in std environments, or performance.now() as u64 in a WASM AudioWorklet).

use dominant_speaker::ActiveSpeakerDetector;

let mut detector = ActiveSpeakerDetector::new();

// Register two participants (timestamp = 0 ms).
detector.add_peer(1u64, 0);
detector.add_peer(2u64, 0);

// Feed audio levels (0 = loud, 127 = silent, per RFC 6464).
// Simulate peer 1 speaking for 2 seconds at 20 ms cadence.
let mut t_ms: u64 = 0;
while t_ms < 2000 {
    detector.record_level(1, 5, t_ms);   // peer 1: active
    detector.record_level(2, 127, t_ms); // peer 2: silent
    t_ms += 20;
}

// Call tick() on a 300 ms timer — returns Some(SpeakerChange) only on change.
if let Some(change) = detector.tick(300) {
    println!("Dominant speaker: peer {}", change.peer_id);
}

See the README for algorithm details, constants reference, and prior art.

Structs§

ActiveSpeakerDetector: Per-room dominant-speaker detector.
DetectorConfig: Tunable constants for the dominant-speaker election.
SpeakerChange: Emitted by ActiveSpeakerDetector::tick when the dominant speaker changes.

Constants§

TICK_INTERVAL: Recommended tick interval matching mediasoup’s production tuning.

Type Aliases§

DefaultDetector: Convenience alias using u64 peer IDs — backward-compatible with v0.1.x.