Vox Transcribe 🎙️

Audio transcription with speaker diarization optimized for Apple Silicon (M-series).

Installation

1. Easy Install (Homebrew) 🍺

Recommended for most users. This installs a pre-bundled version that does not require a Hugging Face token (it ships with all models pre-loaded).

brew tap ma-r-s/tap
brew install vox-transcribe

2. Developer Setup (UV) 🛠️

For users who want to run from source or contribute. Note: This method requires a Hugging Face token for diarization models.

1. Prerequisites: brew install ffmpeg uv
2. Clone & Sync:

   git clone https://github.com/ma-r-s/vox-transcribe.git
   cd vox-transcribe
   uv sync
   

3. Auth: export HF_TOKEN="your_token" (Get one at huggingface.co)

---

Usage

Simply run the command and follow the prompts, or pass a file directly:

# Direct usage
vox-transcribe meeting.mp4

# Using uv (if running from source)
uv run vox-transcribe meeting.mp4

Options

• --model: tiny, base, small, medium, large-v2 (default), large-v3
• --speakers: Exact number of speakers.
• --format: txt (default), srt, json.
• --output: Custom output path.

---

Performance Tips

• Apple Silicon: This tool automatically uses Metal (MPS) acceleration for peak performance on M1/M2/M3 chips.
• Accuracy vs Speed: large-v2 is the most accurate; base is the fastest.
• Memory: If you hit memory issues, use a smaller model or reduce --batch-size.

Documentation

• Architecture - How the hybrid engine works.
• Maintainer Guide (Brew) - How to build and release new versions.
• Full Installation Guide - Detailed setup steps.

License

MIT