Darren spent the top of the show showing off Breeze TTS 2, a brand-new open-source text-to-speech model ported to run locally on Apple Silicon via MLX. The pitch is real: it builds a custom voice just from a text description (“cute southern sorority girl,” “boring French woman with a slight accent”), does full voice cloning from a reference clip, and gives you granular controls — temperature, seed, repetition penalty — that older voice-clone tools never offered. It also chunks and stitches long-form text automatically, so you could feed it an entire audiobook and let it render overnight. Runs entirely on your own machine, no subscription.
Find it on Hugging Face by searching “Breeze TTS 2.”
[Unrelenting #203 1:05]
