RVC-Boss/GPT-SoVITS

Project overview

GPT-SoVITS is a self-hosted application for zero-shot and few-shot voice cloning and text-to-speech. It produces trained models using as little as one minute of voice data and supports cross-lingual inference in Chinese, English, Japanese, Korean, and Cantonese.

Project type

Audio & Speech

Use cases

Meeting Notes · Audio & Speech

Deployment

Python environment

License

MIT

Best for

• Creators, developers, and general users who need to generate text-to-speech audio or train voice cloning models using minimal voice data.

Key capabilities

• Accepts a 5-second vocal sample to perform instant text-to-speech conversion.
• Fine-tunes the model with one minute of training data to improve voice similarity and realism.
• Performs inference in languages different from the training dataset, supporting English, Japanese, Korean, Cantonese, and Chinese.
• Developing pretrained finetuned preset GPT models for better emotion control.
• Capability to mix different TTS models.

Limitations and risks

• Models trained with GPUs on Macs produce significantly lower quality compared to those trained on other devices; CPUs are used on Macs instead.
• Version 3 had an issue with metallic artifacts caused by non-integer multiple upsampling.

Getting started

• Setup difficulty is rated medium. Create and activate a conda environment (Python 3.10), run the provided install scripts for Windows specifying device and source, and manually download pretrained models into the GPT_SoVITS/pretrained_models directory if not using an integrated package.

Evidence and sources

• GitHub project description: 1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
• README: 1. **Zero-shot TTS:** Input a 5-second vocal sample and experience instant text-to-speech conversion. 2. **Few-shot TTS:** Fine-tune the model with just 1 minute of training data…
• README: [](https://colab.research.google.com/github/RVC-Boss/GPT-SoVITS/blob/main…
• Release: 20250606v2pro
• README: Version 4 fixes the issue of metallic artifacts in Version 3 caused by non-integer multiple upsampling, and natively outputs 48k audio to prevent muffled sound (whereas Version 3…

Reviewed at: 2026-07-17T06:12:50Z