filler-removal

原文涵盖的主题

• Filler-word removal
• Step 0 (CRITICAL): get a verbatim transcript
• Step 0.5 (CRITICAL): cross-check word boundaries against silence
• Cutting rules (follow exactly)
• Iteration loop — do not stop until clean
• After the fillers are gone: pacing
• Reference

Full Production Studio note: this skill drives the toolkit in agamm/video-agent. Clone that repo and run your agent inside it for the scripts/paths below to work; the guidance itself is usable anywhere. Goal: cut filler words so the result sounds natural — no clipped consonants, no abrupt audio jumps, no real speech lost. This is an iterative process; do not stop at the first pass. mlx-whisper silently cleans disfluencies by default — it drops most "um/uh" from the transcript, so there's nothing to cut. This was the root cause of a past failure where fillers survived the edit. transcribe is now verbatim by default: it seeds the decoder with a short hesitation prompt (Um, uh, er, hmm...) and conditiononprevioustext=False, so um/uh are kept. Just run: Do not pass --clean here — that restores whisper's filler-dropping behavior, the opposite of what you want. The seed prompt is deliberately narrow (only short hesitations) to avoid over-transcribing real words like "like"/"so". If the user reports fillers that still aren't in the list, listen to the audio span directly to confirm what was actually said before concluding nothing is there. Whisper word timestamps are jittery and unreliabl

这里只展示节选，请以完整原文和当前要求为准。