fix(translation-demo): preserve audio delay after MoQ update

[Qizot]

Summary

Updates the translation demo to the current MoQ APIs while preserving its configured audio/video delay. The publisher now emits continuous Opus audio through silent periods, preventing DTX from draining the viewer's audio cushion while video remains delayed.

• Uses @moq/publish 0.4.4 and @moq/watch 0.5.0, with every @moq/net dependency forced to 0.3.1.
• Migrates connection discovery, capture, encoding, source, decoder, renderer, and sync handling to the new APIs.
• Updates the translation service lock to moq-ffi 0.2.33.

Why DTX is disabled

The viewer expects continuous timestamped audio to maintain its 4.5-second ring-buffer cushion. Opus DTX suppressed packets during silence, allowing the ring to drain and original audio to resume near the live edge while video stayed delayed. Setting usedtx: false keeps that timeline continuous.

Validation

• yarn build
• uv sync --locked
• .venv/bin/python -m compileall -q translator
• Confirmed Publish resolves to 0.4.4, Watch to 0.5.0, Net to 0.3.1 throughout the dependency graph, and moq-ffi to 0.2.33.

Post-deploy monitoring & validation

Publish speech with silent intervals and confirm original audio and video retain the configured delay. Roll back if audio returns near the live edge while video remains delayed.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
moq-demo	Ready	Preview	Aug 19, 2026 3:29pm