Is post-processing net useful?

Kyubyong / tacotron

A TensorFlow Implementation of Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model

Apache License 2.0

1.83k stars 436 forks source link

Is post-processing net useful? #66

Closed zuoxiang95 closed 7 years ago

zuoxiang95 commented 7 years ago

I have trained the full data set with 40 epochs. My mean_loss curve is like this: 7993cbee7ddf967d6070ef2289e0568

But i find the mean loss 1 is very small.

what's the difference between mean-loss1 and mean-loss2?

zuoxiang95 commented 7 years ago

when i use full data to train model, and set sanity_check = False. Then i find the mean-loss1 cannot converge. here is my mean-loss1 curve: 431578059627820642