# mlx-speech

A Python library that runs text-to-speech, voice cloning and speech recognition entirely on Apple Silicon through Apple's MLX framework, sharing one Python API and command-line interface across several open speech models.

mlx-speech is an open-source Python library for text-to-speech and automatic speech recognition on Apple Silicon Macs, with inference implemented in Apple's MLX framework rather than PyTorch, so audio processing stays on the device with no cloud service required. Published model weights download automatically on first use, and local checkpoints can be loaded for fully offline use afterward.

Supported models cover voice cloning and emotion-tagged speech synthesis, multi-speaker dialogue, sound-effect generation from text, audio editing, and automatic speech recognition, each loaded through a shared tts.load or asr.load call or selected by name from the command line. It requires an Apple Silicon Mac and Python 3.13 or newer, and installs through pip.

Website: https://appautomaton.com/mlx-speech/