Fine tuning tacotron2 to new language

[JRMeyer]

>>> kjk11
[August 17, 2020, 9:02pm]

Hi!

I'm currently trying to fine-tune Tacotron2 (which was trained from
LJSpeech originally) for German, but the training takes about an hour
per epoch and the alignment is improving slowly/not at all.

I've been unfreezing the entire model and re-training it, but should I
be only unfreezing the postnet or some other combination of modules?
Please let me know!

[This is an archived TTS discussion thread from discourse.mozilla.org/t/fine-tuning-tacotron2-to-new-language]

[JRMeyer]

>>> nmstoker
[August 17, 2020, 11:12pm]

- normally there isn't any
need for making decisions regarding unfreezing layers, since to
fine-tune you simply follow the instructions here:
https://github.com/mozilla/TTS/wiki/Training-and-Testing#how-to-run
with your updated config file. slash
The only point to bear in mind is that the directory structure changed
in the dev branch recently so the commands given in the wiki need a
minor adjustment for the directories. The flag of slash --restore_path is the
one you'd normally use.

The other thing if it's slow is to double check that you definitely have
it using CUDA and the GPU correctly. The output right after training
starts should confirm this (I had the wrong device ID once and sat there
for a while thinking it was rather slow compared to before!)

I've got to say I haven't looked seriously at fine tuning between
languages, so maybe you're unfreezing has more merit then but I'd
probably start by trying it the way it's documented unless you've been
advised to try this approach here and I've simply missed it.

[Archived Post]

[JRMeyer]

>>> mrthorstenm
[August 18, 2020, 4:55am]

, slash
welcome to our community
I've no idea on training on a language switch, but since you're talking
about a german model maybe this is helpful for you.

Download of my german dataset: slash

{.avatar
tts](https://discourse.mozilla.org/t/contributing-my-german-voice-for-tts/48150/112)
drop-close='true'
.box}

> Dataset 'thorsten' is now available for free to use german tts
> training. See my github page for dataset details and download url.
> Please read „special thanks' section on github for a list of great
> supporters on this project. It's a pleasure to work with you guys on a
> free german tts model.

The model training is still in progress (as you can see in the linked
thread), but maybe that's a point to think about before starting a
cross-language training.

[Archived Post]