Test Drive: e2tts with MLX
Published on .
Today I test drove 2 different implementations of Microsoft’s TTS in MLX https://github.com/lucasnewman/e2-tts-mlx and https://github.com/JosefAlbers/e2tts-mlx. One came with a tiny pre-trained model, the other no pre-training. So the process was:
- Install
- Train
- Synthesize
Well… I haven’t gotten past step 2 yet. So, not much to report. I started off by fighting with Python a bit, since e2tts-mlx was cranky about needing 3.12.3 or higher and my version of pyenv was outdated. So after that, I was able to generate some speech! with e2tts "text here" only to discover that the pre-trained model is not good. So… I tried to train but haven’t completed yet.
For e2-tts-mlx, after installing the requirements, I can run their train_example.py but it never makes any progress or does much (with just 0.6% CPU use)