Transcribes meetings, voice notes and videos into text with timestamps, on our servers, with Danish and English detected automatically.
Say it in your own words; the assistant finds this recipe by itself. Examples:
Turn a recording into text in ONE command, on our servers, no cloud speech service: meetings, voice notes, interviews, podcasts, phone recordings in any common audio or video format (m4a, mp3, wav, mp4, webm, ogg). You get a transcript as Markdown or plain text with timestamps, optionally as subtitles (SRT/VTT); language detected automatically, Danish and English work well. Use for: transcribe this meeting, write up my voice memo, subtitles for this video, what was said in the call, referat fra mødet, transskriber optagelsen. Triggers: transcribe, transcript, recording, voice memo, meeting audio, subtitles, SRT, podcast, interview, transskription, optagelse, lydfil.
python3 /skills/transcribe/scripts/transcribe.py IN OUT [--lang da|en|auto] [--format md|txt|srt|vtt] [--start 0:00 --end 10:00]
converts the file with ffmpeg and runs faster-whisper (the multilingual "small" model, int8,
on CPU). Inputs are in /work/in, results in /work/out. Prints one OK … line (duration,
language, words, seconds taken) or one ERROR … line.
| Need | Command |
|---|---|
| Transcript with timestamps (Markdown) | transcribe.py /work/in/meeting.m4a /work/out/meeting.md |
| Plain text, no timestamps | transcribe.py /work/in/memo.mp3 /work/out/memo.txt --format txt |
| Subtitles | transcribe.py /work/in/video.mp4 /work/out/video.srt --format srt (or vtt) |
| Force the language | add --lang da (auto-detection is the default and is reliable past ~20 s of speech) |
| Only a part | add --start 0:00 --end 10:00 |
| How long is it, which language | transcribe.py probe /work/in/meeting.m4a |
Speed on this server is roughly real time: a 30-minute recording takes 20–40 minutes, so run
anything over 3 minutes with start_job and poll job_status; use sandbox_run only for short
clips or a --start/--end sample. Timestamps are [mm:ss] per segment. Give the person the
download link; summarise the transcript only if they asked for a summary.
SKILL.mdgate.jsonscripts/__pycache__/transcribe.cpython-311.pycscripts/__pycache__/transcribe.cpython-313.pycscripts/transcribe.pyWritten by Percerno, MIT licence.