type: docs
area: subtitles

- Explain how dialogue generation retains uncertain audible sections, why songs may also be transcribed, how detected speech starts guide long-passage cuts to reduce early subtitles, and why each passage uses a fresh Whisper process.
- Document automatic loaded subtitle references, English dialogue preference, signs/songs filtering, fallback behavior, and the limits of reference-guided timing.