GPT-SoVITS-Infer 0.1.1

Inference code for GPT-SoVITS

GPT-SoVITS-Infer

This is the inference code of GPT-SoVITS that can be developer-friendly.

Usage Example

Check out the example notebook for a quick start. Or open it in Colab

Prepare the environment

As we all know, the dependencies of an AI project are always a mess. Here is how I prepare the environment for this project, by conda:

conda install python=3.10
conda install pytorch=2.1 torchvision torchaudio pytorch-lightning pytorch-cuda=12.1 -c pytorch -c nvidia 
conda install ffmpeg=6.1.1 -c conda-forge

You can also try to prepare the environment with cpu only options, which should work, but I have not tested it yet.

After the environment is ready, you can install the package by pip:

pip install GPT-SoVITS

I do not add the packages related to torch to the dependencies of GPT-SoVITS-Infer. Check if the environment is ready if things go wrong.

Advanced Usage

• GPTSoVITSInference.load_sovits and GPTSoVITSInference.load_gpt: You can load your own fine-tuned model by the methods.
• GPTSoVITSInference.set_prompt_audio: Set the prompt audio for the inference.
• GPTSoVITSInference.get_tts_wav_stream: Return a generator that yields the audio pieces of the generated audio. It will create a background thread to generate the audio, so you can get the audio pieces while the audio is still being generated.