The Implementation of FastSpeech Based on Pytorch.
Modified from xcmyz/fastspeech
Add Conformer and UNet Post-Net, which can be called in hparams.py.
This repo does not include alignment and dataset. Pretrained models(201000 steps) are in the google drive link:
https://drive.google.com/drive/folders/13F2oh1gdOTvsYBwyvMuJ34BBCq97wK2c?usp=sharing
Comparison of mels before(original) and after post-net(predicted) is here:
