| name | stt |
| description | Transcribe audio or video files to text locally on macOS using parakeet-mlx (NVIDIA Parakeet via Apple MLX). Use when the user asks to transcribe an audio/video file, get a transcript, generate subtitles/captions (srt/vtt), or convert speech in a file to text. Triggers on "transcribe this", "make a transcript", "get subtitles for", or pointing at an .mp3/.wav/.m4a/.mp4/.mov file with intent to extract speech. Apple Silicon only. |
Transcribe Audio
Local, offline transcription via parakeet-mlx. Requires Apple Silicon and ffmpeg. Resolve both commands with command -v before use and report a missing prerequisite instead of assuming an installation path.
Run
parakeet-mlx <file>
parakeet-mlx <file> --output-format txt
parakeet-mlx <file> --output-format all
parakeet-mlx *.mp3 --output-dir ~/transcripts
Default output format is srt. Pass --output-format txt when the user wants plain prose, not subtitles. Output lands in the current dir unless --output-dir is set.
Notes
- First run ever downloads the model (~600MB) from Hugging Face, then it's cached. Expect a one-time pause.
- Default model
mlx-community/parakeet-tdt-0.6b-v3 is multilingual (~25 languages). Override with --model if needed.
- Accepts audio and video (mp3/wav/m4a/mp4/mov/...);
ffmpeg handles extraction.
- For languages outside Parakeet's set, fall back to
whisper-cli (brew install whisper-cpp), which covers ~99 languages.
