[generate.py] nan value when generating samples

[hamgcho]

I used sc09 dataset for the training dataset.
And I for sure followed the preprocessing code which is given here in the repo.

But as below, when I run generate.py, I am consistently facing the nan values :

The direct error message I found is as below :

I used debugger and found out the rnn_cell of the WaveRNN has the nan values

I want to know

• if you have faced this kind of situation before,
• and if you did, how you resolved the problem.

Thank you for sharing this understandable codebase in public!