NV homework repository of HSE DLA course. The goal of the project is to implement HiFi-GAN model and train it on LJ-Speech dataset.
This homework is aimed at implementing HiFiGAN vocoder. We understand that anyone can find the authors' source code, but please don't write it off mindlessly. (At least the reviewer will see it and penalize you)
Use dataset LJSpeech you already know.
To install necessary python packages run the command:
pip install -r requirements.txt
Download all needed resources (data, checkpoints & inference examples) with
python3 bin/download.py
If you wish to create a Yandex DataSphere Dataset, specify the option
python3 bin/download.py -d
One may use datasphere.ipynb notebook that contains all necessary commands to reproduce the results of the project.
Once the resources are ready, start the training with
python3 train.py
Use
python3 inference.py
to generate test audio files.
The audio is stored at resources/predicted.
Run python3 some_script.py -h for help.