EATS

A pytorch implementation of the EATS: End-to-End Adversarial Text-to-Speech (https://arxiv.org/pdf/2006.03575.pdf)

Attention

I only implemented the preliminary framework.
In my opinion, the training cost of EETS is very expensive. Maybe we can consider using ground truth duration during training process and introduce additional auxiliary loss to accelerate convergence.

Author

Geng Yang (@yanggeng1995) Jian Cong (@npujcong)