Meeting Transcription

Private local meeting transcription for OpenClaw: convert audio recordings into timestamped transcripts, SRT/VTT subtitles, and structured Markdown notes with deterministic action-item extraction. Use for meetings, interviews, calls, lectures, and voice notes when audio should remain local.

Private local meeting transcription

Use meeting.py to transcribe a user-selected audio/video file with local Whisper and produce reusable transcript artifacts. The workflow does not call a cloud transcription API. Whisper model files may be downloaded the first time a model is used; disclose that setup network requirement and keep later processing local.

Quick start

bash

python3 meeting.py --input meeting.m4a --model base --output-dir meeting-output

This writes:

• transcript.json with transcript text, language, duration, and timestamped segments;
• transcript.srt with subtitle timing;
• notes.md with a readable transcript and deterministic action-item candidates.

Formats and models

• Use --format all (default), json, srt, or markdown.
• Use Whisper models tiny, base, small, medium, large, or turbo.
• Use --language en when the language is known; omit it for automatic detection.
• Use --include-word-timestamps only when detailed timing is required; it costs more compute.

Privacy and safety

• The input must resolve inside the current workspace; symlink escapes are rejected.
• Output is written only to the requested output directory inside the workspace.
• The script performs local transcription through Whisper and makes no runtime network request.
• Initial model acquisition can require network access through Whisper's normal model cache. Use a pre-downloaded model in restricted environments.
• Do not place confidential recordings, credentials, or private transcripts in public issue reports.

Agent workflow

1. Confirm the user has permission to process the recording.
2. Check local capacity before large jobs with the Local System Info skill.
3. Choose base for a practical default; use small or larger when accuracy matters and resources allow.
4. Run the script and inspect transcript.json before summarizing.
5. Cite timestamps when extracting decisions or action items.

Example

bash

python3 meeting.py \
  --input interview.wav \
  --model small \
  --language en \
  --output-dir interview-output \
  --format all

Source: https://github.com/iyeque/openclaw-skills/tree/main/skills/meeting-transcription

Report bugs and security concerns through the repository issue guidance.