Turn a long YouTube talk into a local HTML transcript with frames
At a glance
A Lobsters commenter used yt-transcribe and MLX Whisper to create a readable local copy of a talk. This workflow is listed under Transcription, Local AI, Learning Tools.
- Agents / tools
- yt-transcribe, MLX Whisper, yt-dlp, FFmpeg
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Use a Mac with Homebrew and install dependencies: brew install ffmpeg jq python yt-dlp, then pip install mlx_whisper. Download the yt-transcribe file from https://github.com/llimllib/yt-transcribe into your working folder and run chmod a+x yt-transcribe. 2) Run ./yt-transcribe -thumbs -outdir ./talk-notes -outfile talk.html 'VIDEO_URL', replacing VIDEO_URL with the YouTube watch URL. The script downloads media, transcribes it locally with MLX Whisper and creates an HTML page. 3) Open talk-notes/talk.html in a browser and read the transcript alongside the extracted frames. Omit -thumbs if you want text alone. Revisit the audio when a passage needs checking.
How it went
The commenter reported downloading and transcribing the linked talk in two minutes on an M1 MacBook Pro.
Worth knowing
The installation and command options were cross-checked against the author's repository. That timing is one machine's report.
