This project provides an enhanced API for GPT-SoVITS, a zero/few-shot Chinese voice cloning model. It addresses limitations in the original API, such as poor handling of mixed-language input and sentence splitting, offering a more robust interface for developers and researchers integrating voice cloning into applications.

How It Works

The API is built upon the GPT-SoVITS framework, offering improved text processing capabilities. It supports splitting text by punctuation for more natural speech synthesis and allows for mixed-language inputs within a single request. The API can be configured to use default reference audio or accept specific reference audio paths and text during API calls.

Quick Start & Requirements

• Install by downloading api2.py and placing it in the GPT-SoVITS directory.
• Run using .\runtime\python api2.py (or python api2.py on Linux).
• Requires a pre-existing GPT-SoVITS installation.
• Supports CPU, CUDA, and MPS (Mac) inference.
• Official GPT-SoVITS deployment guide: https://github.com/SkillNing/GPT-SoVITS

Highlighted Details

• Supports mixed Chinese-English and Japanese-English speech synthesis.
• Offers configurable text splitting based on punctuation (1-5 levels).
• Allows specifying inference device (CPU, CUDA, MPS) and binding IP/port.
• Enables using custom trained models by specifying their paths.

Maintenance & Community

This is a community-driven enhancement to the GPT-SoVITS project. Further community support and development can be found via the original GPT-SoVITS repository.

Licensing & Compatibility

The licensing details are not explicitly stated in the README for this specific API enhancement. It inherits the licensing of the underlying GPT-SoVITS project, which is typically Apache 2.0, allowing for commercial use and integration with closed-source applications.

Limitations & Caveats

The API does not support dynamic model switching; separate API servers must be launched for different models on different ports. The README does not specify version compatibility with the core GPT-SoVITS project.