Open-source few-shot voice cloning system — clone a voice from 1 minute of reference audio. Runs locally on consumer GPUs, multilingual (EN/JP/ZH/KO). Free, MIT-licensed, self-hostable.