A PytorchLightning implementation of mel-spectrogram vocoder using WaveNet. Created with reference to Chainer implementation.
- Bulid Docker image.
sudo docker build -t pytorch_lightning_wavenet .
- Run Docker container.
sudo docker run --shm-size=512m --gpus all --rm -v $PWD/wavenet:/wavenet -w /wavenet -it pytorch_lightning_wavenet:latest bash
- Download dataset.
wget http://www.udialogue.org/download/VCTK-Corpus.tar.gz
tar -xf VCTK-Corpus.tar.gz
- Start training.
python train.py --dataset <directory of dataset e.g. ./VCTK-Corpus/>
- Generate audio with trained model.
python generate.py -i <input file> -m <trained model e.g. ./lightning_logs/version_0/checkpoints/last.ckpt>