RVC WebUI

RVC WebUI is a local AI voice conversion tool for people who want to train a custom voice, change the voice in a recording, or use a live voice changer. It runs on Windows and Linux, including Ubuntu servers, with a browser interface for training and conversion and a separate interface for live use. It's free and open source under the MIT license.

Its retrieval approach uses examples from the trained voice to reduce traces of the original speaker's vocal tone in the output. You can train with a small dataset; the project recommends clean voice recordings with little background noise. The base models use the VCTK speech dataset, and training can work on relatively modest graphics cards.

Model merging lets you blend trained models to alter a voice's tone. For work with songs, pymss/MSST models separate vocals from accompaniment, while RMVPE extracts vocal pitch for conversion. RVC accepts voice models in .pth format with accompanying .index files.

NVIDIA GPUs use CUDA acceleration. AMD and Intel hardware can use DirectML on Windows; the documented Linux path for that hardware uses the CPU. Training and conversion run on your own machine or server, with model assets available to download from Hugging Face.

The live voice changer supports low-latency audio. Performance depends heavily on hardware and driver support, including ASIO input and output devices.