GPT-SoVITS

GPT-SoVITS

Open-source few-shot voice cloning: train a high-similarity TTS model with just 1 minute of speech

Last Updated Nov 17, 2026
GPT-SoVITS screenshot

介绍

GPT-SoVITS is an open-source few-shot voice cloning and text-to-speech (TTS) project with over 60,000 stars on GitHub, making it one of the most popular voice cloning solutions in the Chinese community. With just about 1 minute of dry voice samples from a target speaker, you can fine-tune a highly similar personalized TTS model; it also supports a zero-shot mode for quick cloning from a few seconds of audio.

Core features include: multilingual speech synthesis for Chinese, English, Japanese, Korean, and more; few-shot fine-tuning for high-fidelity timbre replication; a WebUI integrating the full training and inference workflow; and support for mixed Chinese-English reading with emotion control. The project is open-sourced under the MIT license, with a Windows all-in-one package and Docker deployment options.

It is ideal for creators, developers, and VTuber operators with a local GPU (NVIDIA graphics card), especially users who need highly customized voices, value data privacy, and prefer not to pay per character. Its cloning naturalness for Chinese content ranks among the best among open-source solutions, making it a go-to productivity tool for Bilibili creators and fan-content creators.