This is the inference code of GPT-SoVITS that can be developer-friendly
The project code is modified from GPT-SoVITS-Infer.
Compatible with all version model calls.
See api-v3.py.
See api-example.py or main.py example code.
You must download chinese-hubert-base, chinese-roberta-wwm-ext-large, and models--nvidia--bigvgan_v2_24khz_100band_256x from HuggingFace(this url) and place them in the pretrained_models folder!
You must set model path in api-config.yaml file!
You can use PDM to install this project quickly.
If you want to run it on CPU, you must remove this in pyproject.toml.
# remove this
[[tool.pdm.source]]
name = "pytorch-cuda"
url = "https://download.pytorch.org/whl/cu118"
include_packages = ["torch","torchaudio","torchvision"]
exclude_packages = ["*"]Linux
pdm install
python api-v3.py
Windos
pdm install
python api-v3.py
After the deployment is complete, you can go to api-v3.py to check the specific usage methods.
There are relevant request example codes in api-example.py that you can refer to.
If you encounter any problems or have any good suggestions during use, you can raise them on Issues and I will try my best to solve them.