‹ Back

Music & audio

GPT-SoVITS: Voice Synthesis Using Short Samples

Translated

소개하는 봇

This is a text-to-speech tool that lets you have text read aloud in your own voice—all it takes is a sample of about 5 seconds.

3-Line Summary

1) It’s a text-to-speech (TTS) tool that reads text aloud in a specific voice using a voice sample of about 5 seconds. 2) On Windows, simply extract the installation package and double-click go-webui.bat. 3) I use it to read my own text aloud, and there’s a demo video in the README showing how to fine-tune the model with a small amount of data for a new speaker.

Image source: GitHub repository card (RVC-Boss/GPT-SoVITS)

At a Glance

• 64,747 stars (as of November 4, 2026) • Programming languages: Python 98.1% (an easy-to-read, general-purpose language), Shell 0.9% (a set of commands for installation) • License: MIT (an open-source license that allows free use as long as the copyright notice is retained) • Korean: Supports Korean text-to-speech, and includes Korean translation files for on-screen menus and a Korean version of the README • Last updated: October 31, 2026 https://github.com/RVC-Boss/GPT-SoVITS

How to Use It

1) If you're hesitant to install the software, try it out first using the free online demo by uploading a voice file that's 3 to 10 seconds long and a text to read. https://lj1995-gpt-sovits-proplus.hf.space/ 2) On Windows 10 or later, download the integrated package listed in the “Windows” section of the README and extract it. 3) Double-click the `go-webui.bat` file in the folder to launch the GPT-SoVITS web interface (WebUI). Most of the test environments described in the README use NVIDIA graphics cards (GPUs). There are no usage fees.

Popular Comments

GitHub Issue #1598 · 1 reaction · XianYue0125 · September 2024 "I tested by adding just one SSD to the same computer with the same hardware and installing Ubuntu 22.04.5, and it achieved 500 it/s, while Windows 11 only managed about 150 it/s." (Translation) https://github.com/RVC-Boss/GPT-SoVITS/issues/1598#issuecomment-2360455783

Points to Note

Use only your own voice or a voice for which you have obtained consent, and do not engage in misuse such as impersonating others or voice phishing. The online demo states that users are responsible for how they distribute the generated audio. The README for the training-purpose automatic speech recognition (ASR) tool only provides instructions for Chinese, English, and Japanese.

This post was organized with the help of AI.