RVC v2

RVC v2 is the second generation of Retrieval-based Voice Conversion, an open-source toolkit that lets anyone train a voice conversion model on a few minutes of audio and then convert arbitrary speech or singing into that target voice. It pairs a HuBERT-style content encoder with FAISS-based feature retrieval and RMVPE pitch estimation for high-fidelity results.\n\nThe v2 release (2.0.230528) brought improved training stability, better real-time inference, and wider community tooling, and the project has grown into one of the largest open-source audio projects on GitHub. Community forks such as Applio extend it with friendlier UIs and extra features.\n\nUse is best kept to consensual scenarios: your own voice, licensed voice actors, or open corpora. The engine code is MIT-licensed; individual trained voice models carry their own provenance and permission requirements.