Retrieval-based Voice Conversion

1. Set up ⚙️

conda create -n rvc python=3.10.14
conda activate rvc
pip3 install -r requirements.txt

2. Usage 🚀

First, clone our repository.

git clone [email protected]:jjihwan/Voice-Cloning.git

2.1. Dataset Preparation 🦄

Set model_name and make dataset/model_name directory. e.g. dataset/iu

mkdir -p dataset/iu # change iu as your own model name

1. Download source musics in musics folder (more than 4 songs recommended). Store them as following structure:

   musics
   └── iu
       └── boo.mp3 # names are not important
       └── foo.mp3
       └── moo.mp3
       ...
   

2. Download pretrained vocal-remover from the original repository Download pretrained model

   wget https://github.com/tsurumeso/vocal-remover/releases/download/v5.1.0/vocal-remover-v5.1.0.zip
   unzip vocal-remover-v5.1.0.zip
   mv vocal-remover/models .
   

3. Use vocal-remover to decompose the vocal and instrument.

   python3 vocal_remover.py --input_dir musics/iu --output_dir dataset/iu --gpu 0
   

2.2. Training 🔥

Train your own model

python3 RVC_train.py --model_name iu --save_frequency 50 --epochs 200

2.3. Inference 🎵

1. Prepare the target musics in the target_musics folder.

   target_musics
   └── boo.mp3 # names are not important
   └── foo.mp3
   └── moo.mp3
       ...
   

2. Run vocal-remover

   python3 vocal_remover.py --input_dir target_musics --output_dir target_dataset --gpu 0
   

3. Inference with your own model

   python3 RVC_inference.py --model_name iu --target_dir target_dataset/vocals
   

   You might have to modify L17 to adjust the key if the model and the target keys are different. You can find the results in the results folder.

4. Compose with instruments

   python3 compose_song.py --model_name iu --target_dir target_dataset/instruments
   

2.4. One-click Training & Inference 🤩

Run the run.sh file after preparing the training datasets(in 2.1.1) and target musics(in 2.3.1). You can train the model and inference at once by the following command! 🔥🔥🔥

sh run.sh iu  # change iu as your own model name

Acknowledgement 🤗🤗🤗

Our codes are built on two nice open-source projects, RVC-project and Vocal-Remover. Thanks for the authors!