GPT-SoVITS Voice Training Tool

Train your own custom voice — fully visual, no command line needed.

Free Tool · No Activation Required

Download Official Package

The training tool is built into the KOA launcher, but training requires the GPT-SoVITS runtime environment.
Download the official package (~8 GB) from the links below, extract it, and bind the directory in KOA.

China (Yuque mirror)

Official Yuque docs page with multiple download mirrors

Go to Page

International

Direct download from the official HuggingFace repository

Download .7z

RTX 50 series GPUs (5070 / 5080 / 5090) require the -nvidia50 variant. China users: select the matching version on the Yuque page. International users: download the nvidia50 build here.

Three Steps to Start

  1. 1

    Download & Extract

    Choose your download source above and extract the package with 7-Zip to any location.

  2. 2

    Bind Directory

    In the KOA launcher, go to the Tools page and find "GPT-SoVITS Voice Training Tool". Select the directory you extracted. KOA will auto-detect it and inject the training panel.

  3. 3

    Launch & Start Training

    Click "Launch" and the training panel will open in your browser. A built-in beginner guide walks you through every step.

This tool is built on the open-source project GPT-SoVITS (MIT License) and its community ecosystem. It is a free bonus tool bundled with KOA and is not part of the paid activation scope. Please do not purchase this tool from any channel; voices you train should be used for personal purposes only, in compliance with your local laws and regulations.