What it does
Finds spoken mistakes in a recording: stutters, repeated words, false starts and retakes. A script proposes candidates, then each one is reviewed in context, because emphasis and a stutter can look the same. Only approved cuts are rendered with FFmpeg. It is meant to run after cut-silences.
Try saying
- /cut-mistakes
- Remove stutters and repeats from this video
- Keep the best take
What it needs
- A word-level transcript (JSON)
- FFmpeg
- Node.js
What is in the zip
3 files, 9 KB zipped
- scripts/apply-cuts.mjs9 KB
- scripts/find-cut-candidates.mjs8 KB
- SKILL.md4 KB
Credit
Nate Herk's HyperFrames student kit, MIT