LuxTTS: Open-Source Voice Cloning on a 4GB GPU from Three Seconds of Audio

A new open-source text-to-speech model can clone a voice from just three seconds of audio and runs at 150x realtime on consumer hardware — no cloud subscription required.