Transcript Cleanup

Meeting transcripts and voice notes often contain timing lines, filler words, speaker noise, and one-off context. Sending that raw text straight into a skill generator makes weak sources look stronger than they are.

source-to-skill clean-transcript performs a conservative cleanup before the readiness gate runs.

If the source is audio, transcribe it first with source-to-skill transcribe-audio or another transcription tool, then clean the transcript.

What It Does

• removes SRT / VTT timing lines
• removes common English filler words
• removes a small set of common Chinese filler fragments
• preserves speaker labels when they are written as Name: text
• renders a Markdown file with numbered segments

What It Does Not Do

• it does not summarize
• it does not invent structure
• it does not decide that a transcript deserves a full skill
• it does not replace human review for sensitive recordings

Example

source-to-skill clean-transcript examples/meeting-transcript.vtt \
  --out out/clean-meeting.md \
  --title "Meeting Transcript"

source-to-skill analyze out/clean-meeting.md

The cleanup step should make the source easier to score, not more impressive than it really is.