Released Model - Tacotron2 - Few errors popping up + necessary fixes

TTS (Text-to-Speech)
#1

Hi Guys and @erogol -

This is my first time posting/contributing so not sure if this is the right place to post.

Noticed a few things were outdated or did not align in terms of the older Tacotron model that is released for the files + config.json for the first released model - https://github.com/mozilla/TTS/wiki/Released-Models. Not sure if you want to put a sticky for people who want to reference in the future.

  1. This error is still popping up - fix is here but edits need to be made https://github.com/mozilla/TTS/issues/401
  2. config.json in your checkpoint file is missing a few parameters for some reason in comparison to the older TTS file. Need to incorporate those in and should work.
#2

I know it is bit confusing, I simply use the dev branch, all others can be flawed.

#3

@othiele were you able to get the released models to work with the dev branch? Usually I would get layer errors during training and ruin the current model

#4

This setup works well for German:

https://colab.research.google.com/drive/1SPl226SwzrfMZltrVagIXya_ax4CsMh-?usp=sharing

And the DDC models are the newest - I think, check the English one from the wiki