RVC WebUI

RVC WebUI is a local AI voice conversion tool for people who want to change the voice in recordings or use a live voice changer. It also lets you train a voice model from your own recordings.

• Custom voice training: Train with a small dataset, with clean recordings and little background noise recommended for better results.
• Voice conversion: Use trained voice examples to reduce traces of the original speaker's tone, with RMVPE handling vocal pitch extraction.
• Live voice changing: Convert microphone audio in real time through a separate interface; latency depends heavily on hardware and audio driver support.
• Model blending and vocal separation: Blend trained models to change voice tone, or use pymss/MSST models to separate vocals from accompaniment.

Training and conversion use a browser interface and run on Windows or Linux, including an Ubuntu server without a desktop. Custom voice models use .pth files with accompanying .index files, and model assets are available from Hugging Face.

The project is open source under the MIT license. NVIDIA GPUs support CUDA acceleration. AMD and Intel hardware can use DirectML on Windows; the Linux path for those devices uses the CPU. ASIO input and output devices can reduce live conversion latency when the hardware and drivers support them.