MelGAN

A Pytorch Implementation of MelGAN (Mel Spectrogram --> Waveform)

Requirements

PyTorch 1.2.0 & python 3.6 & Librosa

Examples

Step 1. Download Dataset

• LJSpeech : https://keithito.com/LJ-Speech-Dataset/

Step 2. Preprocessing (Preparing Mel Spectrogram)

python preprocessing.py --in_dir ljspeech --out_dir DATASETS/ljspeech

Step 3. Train MelGAN

-c: configurations & hyper parameteres in json

-m: model directory name

python train.py -c configs/base.json -m test

Step 4. Synthesize

Run Sample Test.ipynb. If it doesn't work, please try nbviewer.

While improving, I share a temporary checkpoint of generator, which only runs about 140K steps: link

Put the checkpoint file into "./logs/test/".

You can get an synthesized audio sample of the temporary generator on the above notebook file.

References

• MelGAN: https://arxiv.org/abs/1910.06711
• Base codes are higly adopted from: https://github.com/ksw0306/ClariNet