This project provides a unified Gradio and React web UI for numerous cutting-edge text-to-speech (TTS) and audio generation models, targeting researchers and power users seeking a consolidated platform for AI-driven audio synthesis. It aims to simplify the integration and experimentation with a wide array of advanced models, offering a single interface for diverse audio generation tasks.

How It Works

The project leverages a modular architecture, integrating various TTS and audio generation models as extensions. It utilizes Gradio for the backend API and React for the frontend, providing a dynamic and interactive user experience. This approach allows for easy addition and management of new models, facilitating rapid experimentation and comparison across different synthesis techniques.

Quick Start & Requirements

• Install: Download the latest version and run start_tts_webui.bat (Windows) or start_tts_webui.sh (macOS, Linux). The script sets up conda and Python virtual environments.
• Prerequisites: NVIDIA GPU with CUDA 12.4 recommended for optimal performance. Docker is also supported.
• Resources: Initial model downloads can be substantial.
• Links: Installation Guide, Docker Setup, Google Colab.

Highlighted Details

• Supports over 20 TTS and audio generation models including Bark, Tortoise, XTTSv2, StyleTTS2, MusicGen, and RVC.
• Features an extension manager for easy model integration and management.
• Offers both Gradio and React UIs for a comprehensive user experience.
• Includes a Docker image for simplified deployment and reproducibility.

Maintenance & Community

• Active development with frequent updates, including recent additions of OpenVoice V2, DIA, Kokoro TTS, CosyVoice, and GPT-SoVITS extensions.
• Discord community available for support and discussion: https://discord.gg/V8BKTVRtJ9.

Licensing & Compatibility

• Core codebase is MIT licensed.
• Dependencies have varying licenses; some, like encodec and diffq, are CC BY-NC 4.0, restricting commercial use. lameenc and unidecode are GPL.
• Model weights also have different licenses; users must verify compatibility for their intended use.

Limitations & Caveats

• Some dependencies (e.g., encodec, diffq) have non-commercial licenses (CC BY-NC 4.0), potentially limiting commercial application.
• Audiocraft extensions are currently Linux/Windows only; macOS support may require manual installation.
• The project combines many AI models, leading to potential dependency conflicts and warning messages during installation, which are noted as expected.