GPT-SoVITS

GPT-SoVITS (RVC-Boss, MIT licensed) is the open-source voice-cloning toolkit that pairs a GPT-style text-to-semantic model with SoVITS synthesis. Five seconds of reference audio gets zero-shot TTS; a minute of data fine-tunes a markedly better voice. It covers English, Chinese, Japanese, Korean, and Cantonese, and ships a WebUI for dataset building, ASR labeling, and training. Honest caveats: this is a tinkerer's toolkit, not a polished product — setup is involved and quality depends heavily on training data; and voice cloning raises consent and misuse questions the project does not gate for you.