Closed 78Alpha closed 1 year ago
It seems that my tortoise patch had issues. Did you have the webui installed before? Then I need to bump up the version of tortoise to force the reinstall.
Yes, I had it installed before. I usually download the new zip and reinstall from that. I looked at the latest version of tortoise and didn't see any mention of the tokenizer_vocab_file. I'm going to try to update it again, a full fresh install.
After manually installing tortoise from github and install pytorch with cuda support, it seems to be fine.
Previously I would use the update.py but it seemed to miss some things. I changed to downloading a zip and install the requirements files, but it seems tortoise never got updated.
I'll mark this as closed. Tortoise test audio generated and audiogen produced a sound file as well on the same install.
Thank you for your time!
What I'll do is add the latest tortoise updates, and I'll also merge the vocab file into main. This repo uses my fork or Tortoise which is almost the same but a little bit different. Then I'll bump the version, which will make the update.py work correctly.
Audiocraft 1.0.0 requires transformers 4.31.0, however, Tortoise TTS only works with 4.29.2 due to a regression in 4.31.0 that makes it throw an unexpected keys error.
In order to get the tortoise code to run I had to comment a couple lines as well.
Due to tortoise not using these.
TypeError: TextToSpeech.__init__() got an unexpected keyword argument 'tokenizer_vocab_file'