Mobile-like HIFI-GAN

Anton Nuzhdin, Coursework, 3d year

This repository contains the code for my course project on improving HIFI-GAN. We introduced a model that is an efficient modification of the original work from this paper, which also delivers better quality.

Pre-requisites

1. Python >= 3.6
2. Clone this repository.
3. Install python requirements. Please refer requirements.txt
4. Download and extract the LJ Speech dataset. And move all wav files to LJSpeech-1.1/wavs

Training

python train.py --config config_4m.json --checkpoint_path --num_disc <amount of discriminators> --factor <factor to divide disc's channels amount>

To train different model versions, just provide the corresponding config.json from the repo

Inference from wav file

python inference.py --checkpoint_file [generator checkpoint file path]

Generated audios will be saved in generated_files One can provide custom save directiry through --output_dir.