Saving parameters

[m-toman]

Just saw your comment "Unfortunately I have to put these settings into params in order to save if anyone knows a better way of doing this please open an issue in the repo"

A popular alternative is something like this: https://github.com/m-toman/tacorn/blob/master/wavernn/train.py#L42-L56

Great results :) and nice approach to automatically reduce r over time, I'm currently doing something similar for multi-speaker training where the model struggles the get a good alignment.

[fatchord]

@m-toman Thanks! Yeah that looks good, much appreciated.