# Transcription

> Use when the user wants to transcribe audio files (MP3, WAV, M4A) to text using MLX Whisper on Apple Silicon

- Skill: `varunr89/transcription` (Agent Skill)
- Install (CLI): `npx skillmds@latest add varunr89/transcription`
- Raw SKILL.md: https://api.skillmd.com/api/skills/varunr89/transcription/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: varunr89 (https://skillmd.com/u/varunr89)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/varunr89/transcription

---


# Audio Transcription

Fast local transcription using MLX Whisper with Apple Silicon GPU/Neural Engine acceleration.

## Requirements

- Apple Silicon Mac (M1+)
- Python 3.11+ with mlx-whisper package
- Model: mlx-community/whisper-large-v3-turbo (auto-downloaded on first use)

## Usage

The script at `${CLAUDE_PLUGIN_ROOT}/scripts/transcribe.py` processes MP3/WAV/M4A files:

```bash
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/transcribe.py /path/to/audio/files/
```

- Processes all audio files in the given directory
- Outputs transcription text to stdout
- Reports timing statistics per file

## Setup

```bash
python3 -m venv .venv
source .venv/bin/activate
pip install mlx-whisper
```