[FEATURE REQUEST] AudioGen model has been released.

[ohmygoobness]

Is your feature request related to a problem? Please describe. Facebook just released their AudioGen - Medium - 1.5B https://huggingface.co/facebook/audiogen-medium

Describe the solution you'd like Addition of the model to the webui.

[gitmylo]

I believe audio-webui technically already supports it, i just have to add the models to the list, i'll add this real quick and test it

[gitmylo]

support has been added

[ohmygoobness]

@gitmylo i had to install git+https://git@github.com/facebookresearch/audiocraft#egg=audiocraft and torchmetrics into the venv to work. git+https://github.com/facebookresearch/audiocraft.git doesn't pull the latest fork.

[gitmylo]

I don't think you need bleeding edge to run it, i ran it on the main branch. But reinstalling audiocraft might be needed since i cannot normally check the version unless i lock it to a specific version, since comparisons only work for pypi packages, and with git packages you can only check if the version is equal to a certain version.

[ohmygoobness]

okay, on other note is this possible https://twitter.com/yoachlacombe/status/1684611004443787289?s=20 to gain more speed

[gitmylo]

okay, on other note is this possible https://twitter.com/yoachlacombe/status/1684611004443787289?s=20 to gain more speed

I had already implemented half precision for bark in https://github.com/gitmylo/audio-webui/commit/150f6120be807767ca4de66797e6e2a339a1685f

The other part isn't possible due to using bark from the github repo, while that uses huggingface transformers

[nekogecko2]

@gitmylo i had to install git+https://git@github.com/facebookresearch/audiocraft#egg=audiocraft

Thanks that worked. I had a error after updating to latest version