Qwen3-TTS for Mac
A local text-to-speech tool that runs Qwen3-TTS models on Apple Silicon Macs through MLX, with voice cloning, voice design and built-in voices, fully offline.
Qwen3-TTS for Mac runs Qwen3-TTS text-to-speech models locally on Apple Silicon Macs, from M1 through M4, using Apple's MLX framework, without any cloud service or API key. It supports voice cloning from a five-second audio sample, voice design from a text description, and nine built-in voices with emotion and speed control.
The project ships both larger 1.7B models for quality and smaller 0.6B models for lower memory use and faster inference, downloaded separately from Hugging Face and placed in a local models folder. Setup uses a Python virtual environment and requires ffmpeg.
Running on MLX lets the models use the Apple Neural Engine and GPU directly, which the project reports gives lower memory use and less heat than running the same models through standard PyTorch.