Fre-GAN: Adversarial Frequency-consistent Audio Synthesis

The repository contains my unofficial implementation of vocoder model Fre-GAN.

Running the model

• To run code in Google Colab, open fregan_experiments.ipynb, run the first cell to clone the repository and then follow the instructions to get data.
• As steps, you will need, first preprocess

python preprocess.py

• Then you can easily train

python train.py -n name --training_epochs 3000

• And of course, evaluation step, before that you need to put your .wav file into /test directory

python inference.py -p ./fg_model/test/fg-3000.pt

---

Google Colab:

---

References

In my implementation there may be constructions from implementations:

• HiFi-GAN
• Fre-GAN