What is RVC (Retrieval-based Voice Conversion)?

RVC (Retrieval-based Voice Conversion) is an open-source framework that takes a recording of one voice and re-renders it in the timbre of another, keeping what was sung or said: the words, the melody, the timing. Its own README calls it "a simple, easy-to-use voice timbre conversion / voice changer framework", and the code is released under the MIT License (checked 29 Oct 2026).

How it works, in one paragraph

You train a model on recordings of a target voice. At conversion time RVC analyses the input performance, tracks its pitch (the README recommends the RMVPE pitch extractor), and, in the step that gives it its name, replaces features of the source voice with the closest matching features from the target's training set "using top1 retrieval" to reduce tone leakage. The pre-trained base model it starts from uses "nearly 50 hours of high quality audio from the VCTK open source dataset". The project also ships a real-time voice-changing GUI. For a longer walk-through, see our AI voice conversion explainer.

RVC vs TTS vs singing synthesis

The practical difference: RVC needs a human performance to exist first. That makes it a production tool (guide vocals, demo a topline in another singer's register, double your own voice) more than a generator. This framing is our summary, not the project's.

The consent and legal line

The software licence covers the code, not the voice. MIT lets you use RVC freely; it gives you no rights over whoever's voice you train it on. Train only on your own voice or a voice you have written permission for, and say so when you release. In the EU, Article 50 of the AI Act requires deployers who generate or manipulate audio that counts as a "deep fake" (content resembling existing persons, objects, places or events that would falsely appear authentic) to disclose that it is artificial; for evidently artistic work that duty is limited to disclosing it "in an appropriate manner". Other places add their own rules on voice and personality rights; our guide to protecting your voice from AI cloning covers the defensive side, and licensing your voice for AI covers the paid side. On covers specifically, read are AI-generated covers legal?.

What we think

RVC is the reason "voice cloning" stopped being a lab demo: free code, a modest base model, and a tool that runs on a home machine. The tech is not the hard part any more; permission is. The useful habit for producers is to treat a voice model like a sample: no clearance, no release. And because RVC only changes timbre, the performance still has to be good. A weak take converted into a famous voice is still a weak take, now with legal risk attached.