transcribelisted

# Transcribe Transcribe audio files to text using mlx-whisper. ## Platform requirement **mlx-whisper runs only on Apple Silicon macOS** (it needs the Metal GPU). Supported setups: - **Apple Silicon Mac** — runs locally; needs `pip install mlx-whisper`. - **multiplai container** — bridges to the macOS host via SSH (see Container Support below). - **Plain Linux / WSL / Intel Mac** — not supported by this skill. Tell the user up front and suggest [whisper.cpp](https://github.com/ggml-org/whisper.cpp) or [faster-whisper](https://github.com/SYSTRAN/faster-whisper) as local alternatives. ## Quick Start Run the transcription script: ```bash ${CLAUDE_PLUGIN_ROOT}/skills/transcribe/scripts/transcribe.sh <audio_file> [output_file] [--override] [--model <model_name>] [--task <transcribe|translate>] [--language <code>] ``` ## Models | Scenario | Model (auto-selected) | |----------|----------------------| | English audio | `mlx-community/whisper-medium.en-mlx-8bit` (default) | | Non-English audio | `mlx-community/whisper-medium-mlx` (auto when --task translate or --language is non-English) | The script auto-selects the right model. Override with `--model` if needed (e.g., `mlx-community/whisper-large-v3-mlx` for best quality). ## Workflow 1. **Identify the audio file** - the user will mention the path in the prompt 2. **Run transcription** using the transcription script ${CLAUDE_PLUGIN_ROOT}/skills/transcribe/scripts/transcribe.sh ## Example Usage English audio (default): ```