Transcribe Summarize — Development skill for Claude Code

transcribe-summarize


**This repository is generated.** `transcribe-summarize` is developed in [kevin-burns/claude-skills](https://github.com/kevin-burns/claude-skills) and mirrored here on every merge, so that anyone with recorded meetings who would rather not upload them can find and install it on its own. **Please raise issues and pull requests [in the source repository](https://github.com/kevin-burns/claude-skills/issues)** — changes made here would be overwritten by the next build.

Transcribe audio on your own machine, on macOS, Windows or Linux. Optionally turn the result into meeting notes and a PDF.

Nothing leaves your computer unless you name a network backend in the command, and when you do, the tool tells you what it is about to send and what it will cost before it sends it.

What it does

in	what happens	out
audio — .mp3, .m4a, .wav, .flac, .opus	normalise, trim silence, decode, filter inventions	transcript .md + .srt + .json
video — .mp4, .mov, .mkv, .webm	ffmpeg takes the audio track; the video is ignored	same as above
an existing transcript — .vtt, .srt from Teams or Zoom	cues rejoined, timecodes dropped, speakers attributed	readable .md, no transcription at all
any of the above	you review, then Claude writes it up	meeting notes .md + .pdf

A Teams or Zoom download is an `.mp4` and works as-is — you do not extract the audio first. And if the platform already transcribed the meeting, skip the audio entirely: `normalise_transcript.py` turns the `.vtt` into prose and the notes step is identical.

# transcribing needs a backend, and the backend comes from --with
uv run --with 'mlx-whisper>=0.4.2' --script scripts/transcribe.py meeting.mp4

# everything else is