In addition to using existing models, users can also train models according to their needs. RVC provides links to related tools, which can be found below the video.

RVC is highly regarded in the field of voice conversion, offering powerful features. With this tool, users can convert one voice to another in real-time, adding more creativity and possibilities to applications such as live streaming and video recording. Whether for entertainment purposes or to achieve more personalized sound effects, RVC provides flexible and high-quality voice transformation solutions for users.

It should be noted that due to the high computer configuration requirements of RVC, ensure the use of high-performance hardware devices and sufficient memory resources to achieve the best user experience and voice transformation effects.

In summary, RVC is a powerful open-source tool based on the VITS voice synthesis system, capable of real-time voice transformation. By downloading and installing the package and model package, and following the guidelines for configuration, users can easily perform voice transformation and train their own models as needed. RVC brings more creativity and personalization options to scenarios such as live streaming and video recording, providing high-quality voice transformation solutions for users.

Features

This repository has the following features

Uses top1 retrieval to replace input source features with training set features to prevent voice color leakage

Can be trained quickly even on relatively poor GPUs

Can achieve good results with a small amount of training data (recommended to collect at least 10 minutes of low noise speech data)

Can change the timbre through model fusion (using the ckpt processing options in the ckpt-merge tab)

• Simple and easy-to-use web interface
• Can call the UVR5 model to quickly separate vocals and accompaniment
• Environment Configuration
• It is recommended to use poetry to configure the environment.
• The following commands must be executed in an environment with Python version greater than 3.8:
• # Install Pytorch and its core dependencies, skip if already installed

# Reference: https://pytorch.org/get-started/locally/

pip install torch torchvision torchaudio

# If you are using Windows and Nvidia Ampere architecture (RTX30xx), you may need to specify the corresponding CUDA version of Pytorch according to #21's experience

#pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117
# Install Poetry dependency management tool, skip if already installed# Reference: https://python-poetry.org/docs/#installationcurl -sSL https://install.python-poetry.org
|

python3 -
# Install dependencies through poetrypoetry installYou can also install dependencies using pip:pip install -r requirements.txtOther Pre-model PreparationRVC requires some other pre-models for inference and training.

You can download these models from our

Hugging Face space

The following is a list of all pre-models and other files required by RVC:

hubert_base.pt ./pretrained ./uvr5_weights If you want to test the v2 version model (the v2 version model changes the feature input from 9 layers of hubert+final_proj to 768 dimensions of 12 layers of hubert and adds 3 cycle discriminators), you need to download ./pretrained_v2

# If you are using Windows, you may need this file, skip if ffmpeg and ffprobe are already installed; ubuntu/debian users can install these two libraries through apt install ffmpeg./ffmpeg ./ffprobeAfter that, use the following command to start the WebUI:

python infer-web.py

To test the v2 version model (the v2 version model changes the features from 9-layer hubert+final_proj 256-dimensional input to 12-layer hubert 768-dimensional input and adds 3 cycle discriminators), you need to download the ./pretrained_v2 extra file.If you are using Windows, you may need this file; skip if ffmpeg and ffprobe are already installed. Ubuntu/Debian users can install these two libraries with apt install ffmpeg../ffmpeg ./ffprobe

After that, use the following command to start the WebUI:

python infer-web.py