GPT-SoVITS

GPT-SoVITS is an open-source voice synthesis application with a WebUI for zero-shot and few-shot text-to-speech, voice conversion, and voice cloning. It can use a five-second vocal sample for instant synthesis or about one minute of training data for fine-tuning, with support for Chinese, English, Japanese, Korean, Cantonese, and related workflows.

The project includes tools for audio separation, dataset segmentation, multilingual ASR, transcription proofreading, model fine-tuning, inference, and speech-speed control. It can be installed on Windows, Linux, and macOS, run through Python and Conda, or deployed with Docker Compose; pretrained models and optional ASR and UVR5 models are downloaded separately.