Skip to content
Get packs

RVC

Open-source tool that converts one voice into another using a short voice recording to train on.RVC-Project/Retrieval-based-Voice-Conversion-WebUI

GitHub Download 2.3.260718
Stars
39k
Forks
5.3k
Licence
MIT
Language
Python
Last update
3 months ago
Type
App

What it is

RVC is a voice conversion framework with a web interface. You train a model on a voice recording (the README recommends at least 10 minutes of clean speech), then use it to make one voice sound like the trained voice. It also includes a real-time voice changer mode and an option to separate vocals from music.

  • Train a voice model on a short set of recordings and convert a song or speech clip to that voice
  • Run the web interface to choose between training, inference, and other operations
  • Use the real-time voice changer GUI with a microphone for live voice conversion
  • Merge two models to blend voice characteristics using the ckpt-merge option
  • Separate vocals from accompaniment using the optional pymss/MSST models

Great for

  • Hobbyists who want to make cover songs or voice-changed clips with a trained voice
  • Developers comfortable with Python who want to experiment with voice conversion
  • People with a clean, low-noise voice dataset of about 10 minutes or more
  • Users with an NVIDIA GPU or willing to run on CPU

Not ideal for

  • Anyone needing a no-setup tool: installation involves Python 3.12, PyTorch, and downloading several model files
  • People who want to clone voices without consent, since the README does not address rights or ethics beyond saying the base model is copyright-clear
  • Users needing guaranteed low latency: the README reports about 170ms end-to-end, and 90ms only with ASIO devices that depend on driver support
Step-by-step guide: how to use RVCInstall, first run, examples, tips and FAQs
OpenVoice vs RVC

How to get it

  1. 1

    Clone and build from source

    You'll need: Python 3.12 x64 · FFmpeg (installed via apt on Ubuntu or placed in the project folder on Windows)

    Create and activate a virtual environment (Ubuntu)

    python3.12 -m venv .venv

    Install CPU/AMD/Intel dependencies

    python -m pip install -r requirments_cpu_py312.txt

    Download required models from Hugging Face

    hf download lj1995/VoiceConversionWebUI --revision main --include "hubert_base/*" --local-dir assets

    Start the WebUI

    python webui.py
  2. 2

    Run it

    • Install Python 3.12 and FFmpeg, then create a virtual environment in the repository folder
    • Install the requirements file that matches your hardware
    • Download the required models into the assets folder as described in the README
    • Run python webui.py and open the web interface on port 7865
    • Place your .pth model in assets/weights/ and any .index file in assets/indices/
  3. 3

    Get the most out of it

    • Use at least 10 minutes of low-noise speech when training, as the README recommends
    • Use the ckpt-merge option in the ckpt processing tab to blend models
    • Use the optional pymss/MSST models to separate vocals before conversion
    • On Linux servers without a desktop, run python webui.py --noautoopen
    • Use ASIO audio devices for the lowest real-time latency, if your hardware drivers support them

Downloads

Source code (.zip)main branch · straight from GitHubRelease 2.3.260718Release notes and source

Strengths

  • Offers a web UI with options for training, inference, and real-time conversion
  • Documented to train with a small dataset (about 10 minutes recommended)
  • MIT licensed with pretrained models hosted on Hugging Face

Trade-offs

  • Setup is multi-step and hardware-specific (separate requirements for CPU, NVIDIA RTX 50 series, and older NVIDIA cards)
  • The README is partly in Chinese, and the English docs are linked separately, so some details may be harder to follow
  • Latency and performance depend heavily on hardware and drivers

How it compares

ProjectStarsTypeLevelUpdatedBest for
Retrieval-based-Voice-Conversion-WebUI (this)39kAppintermediate3 months agoHobbyists who want to make cover songs or voice-changed clips with a trained voice
OpenVoice38.1kAI model / agentadvanced1.5 years agoDevelopers who want to build voice cloning or text-to-speech features into their own projects
GPT-SoVITS64.6kAppintermediate5 days agoHobbyists who want to experiment with cloning a voice they have permission to use
Amphion10.3kLibraryadvanced8 months agoResearchers who want reproducible implementations of speech and audio generation models
Filee20.5kAppbeginner4 days agoPeople who regularly convert images or documents and want a drag-and-drop workflow
assistant-ui12.9kLibraryintermediate4 days agoReact or Next.js developers adding an AI chat panel to an existing app

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.