About training speed and reason why using guided attention on 2 heads and 2 layers of Enc-Dec alignments

Hello, thanks for your contributions. This source code is very helpful to me and my colleagues.

I got 2 questions while using this source code.

1. How much you spent on training LJ Speech dataset?
2. Any reason that you applied guided attention to 2 heads and 2 layers of Enc-Dec alignments?


Thank you.