Skip to main content

​
Requirements

  • Apple Silicon Mac — M1, M2, M3, or M4 chip
  • macOS — any version supported by your chip
  • Python — 3.10, 3.11, 3.12, or 3.13
whisper(ml)x requires Apple Silicon. MLX inference does not run on Intel Macs or other platforms.

​
Install whisper(ml)x

This installs all required dependencies, including:
  • mlx-whisper — ASR inference on Apple Silicon
  • torch and torchaudio — CPU wheels, used by VAD and alignment models
  • transformers — wav2vec2 alignment models
  • pyannote-audio — VAD and speaker diarization
torch and torchaudio are installed as CPU wheels from the PyTorch CPU index. They are used for VAD and alignment — not for Whisper inference, which runs via MLX.

​
Model downloads

Models are downloaded from Hugging Face on first use and cached at ~/.cache/whisper by default. You do not need to download anything manually. Short model names are automatically resolved to their mlx-community equivalents: You can also pass a full Hugging Face repo ID directly:
To use a custom cache directory, pass --model_dir on the CLI or set download_root in load_model().

​
Speaker diarization setup

Speaker diarization requires a Hugging Face access token and acceptance of the pyannote model agreement.
1

Create a Hugging Face access token

Go to https://huggingface.co/settings/tokens and create a token with read access.
2

Accept the model agreement

Visit the pyannote/speaker-diarization-community-1 model page and accept the user agreement while logged in to your Hugging Face account.
3

Pass the token to whisper(ml)x

Provide your token via the CLI flag or the Python API:
Without a valid token and accepted model agreement, diarization will fail with an authentication error when downloading the pyannote model.

​
Supported Python versions