GPT-SoVITS GPT-SOVITS Voice Cloning Software: Installation Guide, Error Explanations, and Usage Tutorial

First, thanks to the expert AI探索与发现 for the tutorial video.

One more thing: in the part about editing text and adding annotations, after editing the text you must first click the 【Submit Text】 button.

Computer Configuration Requirements
Installing this software requires some hardware support: a dedicated GPU is required, 8 GB of VRAM is recommended, and it works best with NVIDIA cards. 4 GB can also work, but it is very slow. The OP's is 4 GB, which is too slow. If you don't know how much you have, press Ctrl+Shift+Esc on the keyboard, click Performance in the upper-left, and scroll down on the right to see it.




Software Download Address

This is the project software's introduction page: https://github.com/RVC-Boss/GPT-SoVITS   Click docs in the list, then find Chinese, scroll down on the right to find the Windows package download.
Never mind, I'll just give you the specific address directly: https://github.com/RVC-Boss/GPT-SoVITS/tree/main/docs/cn  Then click the package shown in the image to download; it's more than 4 GB.



Environment Configuration

Install dependencies: https://pan.baidu.com/s/19jQBrTK3jvC6V255ZT30Wg?pwd=tswc  Extraction code: tswc
Be sure to install them in the order the OP wrote for you, otherwise they won't install.



When installing the third one, note that you should only select [Desktop development with C++] in it; it requires 24 GB of C drive space.

Manually Installing GPT-SOVITS
A reminder: if your GPU isn't especially old, failures are environment issues. Manual installation basically won't solve the problem. Try to debug/adjust the versions of the environment software. If you need a model trained for you, you can post a message.

Manual Installation Commands
Open the directory where you want to install, for example the D drive, then type cmd in the path bar. After it opens, run the following commands. This part is covered in the second half of the video; just refer to it..

git clone   https://github.com/RVC-Boss/GPT-SoVITS
cd GPT-SoVITS
python   -m   venv   venv
call    venv\Scripts\activate.bat
pip install torch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

One-click Run Script

@echo on
setlocal
call .\venv\Scripts\activate
python webui.py

Notes:
For installation on MAC, refer to the following link:  https://github.com/RVC-Boss/GPT-SoVITS/issues/165
Colab script attached: https://colab.research.google.co ... xG9neCd?usp=sharing

Model Training Parameters
sovits
4 GB VRAM batch_size: 1, epochs: 8; it can succeed but is very slow. Audio length within 10 seconds; shorter is recommended.
8 GB VRAM batch_size: 2, epochs: 8 or more; audio length within 10 seconds.
The rest doesn't matter much.

The GPT model matters even less; if the above succeeds, there will basically be no problems below.

Explanation of Various Errors
Errors are basically of the following two types:
1. Dependency issues. Install the environment dependencies in the order I said, and it's not guaranteed that all computers are compatible. You can mix and match different versions, for example all the highest versions. Most errors are caused by version incompatibilities among dependencies.
2. GPU issues. There are also many GPU errors. If your VRAM is relatively small, in addition to paying attention to the settings above, you also need to set the virtual VRAM size to give maximum support.

These two types of problems are the most common. If they still can't be solved, they are other minor issues. As long as the GPU is fine, everything else is not a problem. Also, generally just install the integrated package; there's no need to do it manually.
Finally, if it still doesn't work, use Google Drive; it can be used for free.

Lastly, let me tell everyone: as long as your computer configuration is OK and not too old, it can be installed. But with low VRAM, training is very slow, and the trained quality is also poor. For this, I recommend using the cloud online; anyway, it's free.