[BUG REPORT] Tortoise-v2 TTS model doesn't load about audio-webui

[gitmylo]

Could this be the fix for it?

coqui-ai/TTS#2697

it might be, try activating your virtual environment and running
pip install git+https://github.com/coqui-ai/TTS@dev --upgrade (not entirely sure if you'll need upgrade or pip uninstall TTS and then installing the git version)

[gitmylo]

the coqui TTS is just for general generations, so no, it does not have configs.

I am also unable to reproduce that error. I did see someone say they wanted to use tortoise v2 and had the same issue. not sure where the flag was that they were talking about

[Ph0rk0z]

It's right where that error is. The strict=False has to be passed when loading the model. It maybe a problem with the coqui repo. I manually edited the file for myself. Does TTS python expose that so it can be set as a parameter?

In any case, tortoise here only uses one male voice. I even swapped my trained model for it and then it would do 1/2 in the trained voice and 1/2 in the voice of the dude. Tortoise would have to be done on it's own for max effect so I usually use mr_q ai voice cloning for that unless the TTS voice doesn't matter.

[gitmylo]

i don't know, as i have no issues loading tortoise v2 in coqui. i'm on windows 10, what OS are you on?

[Ph0rk0z]

Linux Mint based on ubuntu 22.04

[nekogecko2]

I'm using windows 11. But the same error happens when I run it with "linux" through WSL.

[gitmylo]

whoops, i misclicked

[gitmylo]

I'm not really sure why it works on my pc, but not on other people's pcs, as i have a clean install of TTS, and i didn't modify anything. yet it loads fine for me

[Ph0rk0z]

Probably windows doesn't enforce strict dictionary loading.

[gitmylo]

but nekogecko2 is on windows, so that isn't it

[Ph0rk0z]

Different versions of TTS? Would have to go through coqui github and see what happened with that. It's a problem for LLMs too. Dict slightly doesn't match and then it doesn't load at all.

[gitmylo]

I'm on version 0.14.3, which is currently the latest version on pypi

[Ph0rk0z]

I think me too since I d/l it fairly recently. Back to square one.

[nekogecko2]

Could this be the fix for it?

coqui-ai/TTS#2697

[Ph0rk0z]

Makes too much sense. New transformers enforced more strictness. Instead of setting strict false they.. removed the extra keys.

[gitmylo]

¯\_(ツ)_/¯

[Ph0rk0z]

Probably we were both using newer transformers which is why you didn't get this error.

[gitmylo]

Probably we were both using newer transformers which is why you didn't get this error.

That's still confusing to me, as I had deleted my venv and completely reinstalled before, and still didn't have the error, weird.

[Ph0rk0z]

Well I will try to update TTS and see if it comes back.

[nekogecko2]

it might be, try activating your virtual environment and running
pip install git+https://github.com/coqui-ai/TTS@dev --upgrade (not entirely sure if you'll need upgrade or pip uninstall TTS and then installing the git version)

Thanks! Uninstalling TTS first and then running

pip install git+https://github.com/coqui-ai/TTS@dev" worked!

For any other complete noobs like me, you "activate venv" by drag and dropping the "activate.bat" file to command prompt from \audio-webui\venv\Scripts\activate.bat