Karaoke Video

Composite pipeline:

1. Extract audio from the source.
2. Run Demucs to separate stems (vocals, drums, bass, other).
3. Mix a karaoke audio bed: instrumental at full level + lead vocal at user-chosen attenuation (-12 dB default; mute removes entirely).
4. Build the lyric track:
   • If the user supplied an LRC or SRT, use it.
   • Otherwise, run whisper on the isolated vocals stem (much cleaner transcript than mixed audio) and convert to ASS with per-line timing.
5. Render the karaoke video — original picture + new audio bed + animated subtitles with music-note bullets.

Procedure

1. Resolve tools