# Transcribe Audio

A Claude Code skill for transcribing audio files with speaker diarisation and automatic speaker name identification.

- **Parakeet + FluidAudio** (default): Fast, local transcription with speaker identification. Runs entirely on Apple Silicon (~5 min for 1 hour of audio).
- **AssemblyAI** (optional): Cloud-based, for non-English audio or when explicitly requested.
- **Speaker names**: Automatically identifies speaker names from filename, YouTube metadata, and transcript content.

## Documentation

- See [SKILL.md](./SKILL.md) for complete documentation and usage instructions.
- See [REFERENCE.md](./REFERENCE.md) for detailed reference notes (including filler-word cleanup behaviour).

## Installation

```bash
# Run install
npx skills add HartreeWorks/skill--transcribe-audio

# When asked "Which agents do you want to install to?", select "Claude Code"
# in addition to the default "Universal" list.
```

If you get "command not found", [install Node](https://github.com/HartreeWorks/skills/blob/main/how-to-install-node.md) then try again.

## About

Created by [Peter Hartree](https://x.com/peterhartree). For updates, follow [AI Wow](https://wow.pjh.is), my AI uplift newsletter.

Find more skills at [HartreeWorks/skills](https://github.com/HartreeWorks/skills).