Skip to content

PLAN — Reusable narration audio

Initiative: INIT-VAC-001

  1. Cache first. The content-addressed cache fixes stale audio for existing users and is the foundation for everything else.
  2. IDs and manifest. Add voiceover IDs and write the manifest from existing video commands.
  3. Narration export. Add the narration set types, schema and vac narration tts.
  4. Word timings where providers return alignment (ElevenLabs first).
  5. Docs and release.

Testing uses a fake TTS provider that returns deterministic audio and durations, so cache hits, misses and manifest contents are unit tested without API keys.

The first consumer is PIDL (INIT-PIDL-005), which exports narration sets from protocol walkthroughs and plays the clips in its viewer.