rvc-project/retrieval-based-voice-conversion-webui

Retrieval-based Voice Conversion WebUI (RVC) is a Python tool for changing the voice in an audio recording to sound like a different person. The core problem it solves is voice timbre leakage, when you train an AI to convert voice A to voice B, parts of voice A's characteristics often bleed through into the output. RVC uses a retrieval-based approach to avoid this: rather than purely generating the target voice from scratch, it searches through a large index of reference audio features to find the closest match, producing a cleaner, more faithful conversion. The tool is designed to work with very small amounts of training data. You can train a voice model using less than 10 minutes of audio from the target speaker. Training runs on consumer GPU hardware (NVIDIA cards via CUDA), Apple Silicon (CoreML), or CPU. A trained model can then convert any input audio to the target voice. RVC also supports real-time voice conversion with low latency, approximately 170 milliseconds, making it usable for live applications such as voice chat or streaming. The architecture is based on VITS (Variational Inference with adversarial learning for end-to-end Text-to-Speech), which is a neural network model designed for high-quality speech synthesis. A developer, researcher, or content creator who wants to convert speech audio to a target voice, for use in creative projects, dubbing, voice cloning experiments, or real-time applications, would use RVC. Training requires a GPU; the interface is a web UI (via Gradio) that makes the process accessible without writing code. The primary language is Python.