PyTorch-Voice-Conversion

Implementations of Voice Conversion models.

Method	Paper	Official Code	State
AdaIN-VC	Chou et al.	GitHub
AGAIN-VC	Chen et al.	GitHub	done
AutoVC	Qian et al.	GitHub
VQVC+	Wu et al.	GitHub

Preprocess

• Create a csv file for preprocessing. See data/example_csv/vctk.csv

python gencsv.py vctk -i ~/VCTK-Corpus/ -o data/example_csv/vctk.csv

• Preprocess using a csv file. See data/example_csv/vctk.csv

python preprocess.py mel-trim -i data/example_csv/vctk.csv -o data/feature/vctk

• The output directory structure

data/feature/vctk
└── mel-trim
    ├── p225_p225_001.npy
    ├── p225_p225_002.npy
    ├── p225_p225_003.npy
    ...

Train

Edit the config file and then run the script

python train.py --config config/AGAIN-VC.yaml

Inference

python inference.py --config config/AGAIN-VC.yaml -s <source_wav> -t <target_wav> -o <output_wav_or_dir>