Investigate enabling Opus DTX for voice sources

[kixelated]

Background

We recently added a kind hint (voice | music | auto) to the publish-side audio Encoder (js/publish/src/audio/encoder.ts) that maps to Opus application + signal. We deliberately left usedtx off for now.

DTX (Discontinuous Transmission) lets Opus send sparse comfort-noise frames during silence, which can meaningfully reduce bandwidth for microphone sources. WebRTC ships DTX off by default and only enables it via SDP usedtx=1, so it's not a slam-dunk default — but for our kind: 'voice' path it's worth testing.

Concerns to validate

1. Frame timing. With DTX, the encoder stops emitting frames during silence (or emits one comfort-noise frame every ~400ms). Our watch-side jitter buffer + PLC needs to tolerate that gap without false "stalled" signals or audible artifacts.
2. Comfort noise repetition. When the receiver hasn't gotten a frame in a while, does the decoder hold the last comfort-noise envelope, or do we need to feed it repeated CN frames manually? Worth checking the WebCodecs AudioDecoder behavior for Opus DTX gaps.
3. MoQ frame-per-group model. Each audio frame opens a new QUIC stream. DTX reduces frame rate during silence, so the per-stream overhead is less of a concern, but the relay's handling of long inter-frame gaps on a track should be confirmed.
4. Catalog jitter field. js/hang/src/catalog/audio.ts advertises the max jitter before the next frame. With DTX this can be much larger during silence. Decide if we re-advertise during silence or just bump the static value.

Implementation sketch

• Extend toEncoderConfig to set usedtx: true when kind === 'voice'.
• Add a watch-side capture/playback test with a known silence interval and verify no glitches.
• Measure bandwidth savings on a typical voice call.

References

• WebCodecs Opus registration
• Opus DTX (RFC 6716, §2.1.9)

[coderabbitai]

🔗 Related PRs

#811 - Add Opus decoder [merged]
#1220 - Add jitter tracking to video codecs and catalog metadata [merged]
#1255 - Add dynamic RTT-based jitter mode for adaptive buffer sizing [merged]
#1362 - Add audio encoder reconfiguration [merged]
#1414 - audio: send each frame as its own group [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!