Unofficial implementation of WaveGlow model.

• Gleb Maslyakov

• Sergey Nikolaev

• Nikita Yudin

Clone this repository and initialize submodule, install project requirements:

    git clone https://github.com/neyudin/wavenetglow.git
    cd wavenetglow
    git submodule init
    git submodule update
    pip install -r requirements.txt

1. Download LJ Speech dataset and unpack it to repository root.

2. Set training configuration in file.

3. Start training process:

    python train.py -c config.json

The whole training process can be monitored via Tensorboard while executing the following command (<log_dir> — directory with saved logs for Tensorboard, log_dir parameter in configuration file stands for it, <port_num> — port number to watch training information via https://localhost:<port_num>):

    tensorboard --logdir <log_dir> --port <port_num>

1. Train WaveGlow model from scratch on data.

2. Download pretrained Tacotron2 model.

3. Run inference demo in notebook.