VoRAS (Vocos with Retrieval and self-Augmentation for Speechs)

(This repository is currently in alpha. After completing the training of the pre-training model, it will be a beta version.)

Welcome to VoRAS. VoRAS is a model derived from RVC for fast and lightweight real-time voice change in Japanese.

VoRAS is based on RVC, replacing the vocoder with Vocos and rewriting the overall model architecture into a modern structure.

Note

VoRAS has stopped development. This is because no further improvement in performance is expected. If you want to try it on top of that, please fork this library.

Launch

Windows

Double click webui-user.bat to start the webui.

Tested environment: Windows 10, Python 3.10.9, torch 2.0.0+cu118

Troubleshooting

error: Microsoft Visual C++ 14.0 or greater is required.

Microsoft C++ Build Tools must be installed.

Step 1: Download the installer

Download

Step 2: Install C++ Build Tools

Run the installer and select C++ Build Tools in the Workloads tab.

Credits

• liujing04/Retrieval-based-Voice-Conversion-WebUI
• charactr-platform/Vocos
• ddPn08/rvc-webui
• NVIDIA/BigVGAN
• /rinna/japanese-hubert-base
• teftef6220/Voice_Separation_and_Selection
• あみたろの声素材工房
• つくよみちゃん
• 刻鳴時雨ITAコーパス読み上げ音声素材
• れぷりかどーる