| name | gemini-transcribe |
| description | Transcribe audio files using the Gemini API with an optional custom prompt. Use when asked to transcribe speech, diarize audio, extract meeting notes, summarize recordings, or process audio with specific instructions. Supports any Gemini model that accepts audio input. Default behavior is speaker-diarized transcription; pass a custom prompt to change the output format or task. |
| argument-hint | <audio-file> [--model MODEL] [--prompt PROMPT] |
| allowed-tools | Read, Write, Edit, Glob, Bash(python3 skills/gemini-transcribe/scripts/gemini_transcribe.py *), Bash(ffmpeg *) |
Gemini Audio Transcription
Transcribe audio with speaker diarization via the Gemini API. Zero dependencies beyond Python 3 stdlib.
Script location
skills/gemini-transcribe/scripts/gemini_transcribe.py
Basic usage
python3 skills/gemini-transcribe/scripts/gemini_transcribe.py <audio-file> [--model MODEL] [--prompt PROMPT]
Transcript is printed to stdout. Redirect to save.
Examples
python3 skills/gemini-transcribe/scripts/gemini_transcribe.py recording.m4a
python3 skills/gemini-transcribe/scripts/gemini_transcribe.py recording.m4a --prompt "List every action item mentioned in this meeting."
python3 skills/gemini-transcribe/scripts/gemini_transcribe.py recording.m4a --model gemini-2.5-flash
python3 skills/gemini-transcribe/scripts/gemini_transcribe.py recording.m4a --model gemini-3.1-flash-lite-preview
python3 skills/gemini-transcribe/scripts/gemini_transcribe.py recording.m4a > transcript.md
Supported formats
Any format Gemini accepts: .m4a, .mp3, .wav, .ogg, .flac, .aac
For large files (>20MB), compress first:
ffmpeg -i input.wav -ar 16000 -ac 1 -c:a libopus -b:a 32k output.ogg -y
Available models
| Model | Notes |
|---|
gemini-3-flash | Default, latest gen |
gemini-2.5-flash | Previous gen, good quality |
gemini-3.1-flash-lite-preview | Cheapest, fast |
Environment
GOOGLE_API_KEY must be set. Get one from https://aistudio.google.com/apikey
