How can I install this repo as the default TTS?

[Sotiris-Bekiaris]

For example I want to use this

from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts
from TTS.utils.generic_utils import get_user_data_dir
from TTS.utils.manage import ModelManager

what should my requirements.txt look like in order to install your forked version? Because I am trying to use XTTS 2 from the original repo and I get an error that it seems you already fixed it.

I am talking about this one.

Now I am setting up a conda env but I cant seem to install correctly your repo.

I did this but I am not sure it is correct

pip install git+https://github.com/idiap/coqui-ai-TTS.git -vv

and I get this after I run my code

    from TTS.utils.generic_utils import get_user_data_dir
ImportError: cannot import name 'get_user_data_dir' from 'TTS.utils.generic_utils' (/opt/anaconda3/envs/xtts/lib/python3.9/site-packages/TTS/utils/generic_utils.py)

[Sotiris-Bekiaris]

For context I run

pip list

and I can see that these are installed

coqui-tts               0.24.1
coqui-tts-trainer       0.1.4

also this is what I am getting from the original repo. (this is where I found your repo from as well)

[Sotiris-Bekiaris]

I feel like the fix that was merged in the dev branch here has not yet been released since I can see that the latest release is at May 29 and includes only this workaround.

Am I missing something?

Shouldn't it work though with the workaround?

Anyways I think I will try to find a way to install the HEAD of dev branch, see if that works...

[Sotiris-Bekiaris]

    from TTS.utils.generic_utils import get_user_data_dir
ImportError: cannot import name 'get_user_data_dir' from 'TTS.utils.generic_utils' (/opt/anaconda3/envs/xtts/lib/python3.9/site-packages/TTS/utils/generic_utils.py)

Alright I took a look in your code and I saw that you use the get_user_data_dir like that:

from trainer.io import get_user_data_dir

replacing that import worked and resolved my issue, but now I get once again this error:

TypeError: isin() received an invalid combination of arguments - got (test_elements=int, elements=Tensor, ), but expected one of:
 * (Tensor elements, Tensor test_elements, *, bool assume_unique = False, bool invert = False, Tensor out = None)
 * (Number element, Tensor test_elements, *, bool assume_unique = False, bool invert = False, Tensor out = None)
 * (Tensor elements, Number test_element, *, bool assume_unique = False, bool invert = False, Tensor out = None)

I am sorry if this is getting a bit chaotic.

Here is a screenshot of my pip list just to be sure

[eginhard]

Hey @Sotiris-Bekiaris ! Looks like you already figured out most things.

• Yes, the import should now be from trainer.io import get_user_data_dir (there is some duplicate code in the TTS and Trainer packages that I'm cleaning up to make maintenance easier)
• The code in the current dev head (which you're already getting with your pip install git+...) with the fix from Fix XTTS streaming for transformers update #46 only works for transformers==4.42.* at the moment, so you could manually install a matching transformers version.
• You can follow [Feature request] Compatibility with transformers>=4.43.2 #65 to get updated when a fix for transformers>=4.43.* is available. There is a patch posted there, but I haven't had time to test and merge it yet. I'll make a new proper release after that as well.

[Sotiris-Bekiaris]

Thank you so much for the answer! Installing transformers==4.42.4 solved it. Also thank you for the contribution to OSS!