Question: How do I upgrade the installation?

[ShmuelRonen]

I installed Bark Infinity a few weeks ago. How do I upgrade the installation to the changes you made over time? The version on GradIO.

[ShmuelRonen]

I installed on windows 10 pro. Nvidia RTX 3090

[JonathanFly]

It should be this, run the commands from the same play you run python bark_webui.py. Oh and activate the environment.

git pull
mamba env update -f environment-cuda.yml --prune
python -m pip install --upgrade pip
pip install --upgrade setuptools 
pip install -r requirements_conda_missing.txt

However I changed the name of the environment removing oneclick (since it's not... yet) so you have to edit the .yml file here:

Change that to bark-infinity-oneclick (or whatever your current name is). The --prune command basically just updates, instead of creating new. I'm having trouble with one specific library needed for cloning, it's this one (in the requirements_conda_missing.txt)

pip install fairseq

I tested this and didn't work. It worked for me on on Windows 10, but wouldn't install in a different machine, in Windows 11. I'm not sure if the 11 versus 10 is the difference but but I'm still trying to figure how to make it work automatically. If it works or doesn't, let me know. Unfortunately fairseq can not be installed in the .yml file with mamba because the version is too out of date. Well specifically the windows version only. So the application will work if you mamba install fairseq but the voice cloning features won't work.

[JonathanFly]

BTW, if you are comfortable changing environments yourself, you could try environment-cuda-nightly.yml and it might be faster. Uses latest versions of everything. But it's sometimes unstable.

[ShmuelRonen]

Thanks! it's works for me.

[JonathanFly]

Thanks for letting me know. Maybe it's just my machine, that would be ideal.

[ShmuelRonen]

On another subject. I would like to draw your attention to the fact that if you run the same sentence more than once, the result that appears below for listening, is not the last version... That means I have to go into the results folder and select the latest version to hear it.

[JonathanFly]

Actually it shouldn't have worked, the missing.txt file should be

encodec
rich-argparse
gradio
fairseq
audiolm_pytorch

and audiolm_pytorch has to removed from the .yml. (I'm trying to sort out the install issues at this very moment.)

Did you change anything? Or maybe are you on a different branch? git status does it say 'on branch main or something else? Or if you did something different by chance and that let fairseq install perfectly, it would be good to know.

On your last question, you mean if you multiple iterations? A big limitation of Gradio is you can't have a changing number of audio elements. I think in the short term I'll just a screen with 10 of them so you can at least your last 10 samples easily.

[ShmuelRonen]

I used:

encodec==0.1.1 rich-argparse==1.1.0

[JonathanFly]

Well I was hoping that would make it work in Windows 11, but doesn't help. Worth a shot, thanks.