Some parts sound more better than others.

oh nice you got it working?

Yes.

hmm the samples on the demo page sounded better

I just tried it with the sample on the demo page

Downloading and Extracting Packages
torchvision-0.10.1   | 7.7 MB    | ############################################################################ | 100%
pytorch-1.9.1        | 911.9 MB  | ###########################################################################9 | 100%

PLEASE dude. ITS BEEN ALMOST AN HOUR

hmm have you tried installing the package in colab

took me a couple minutes

to do the whole thing

ok deps done time to pip install voicefixer

im running this locally because colab is mean to me and i have a gpu decent enough to train a whole model in tt2 so it should be fine

oh ok I am running on cpu in colab

i think it thinks that sibilants are noise and filters them out (thats not good) so far the results have not been very impressionable im goign to bed

ive been getting really cool results with this, this can be game changing for some voices thank you for sharing

nvm, sorry for ghost ping.

it's about the half life train ride voice dataset or something

just testing out the voicefixer on hl1 train ride voice.

Try it out guys! https://colab.research.google.com/drive/1dilM9aZ2ljHoktYggTB6uCgjr268O862?usp=sharing

too late, model's done i was wrong about the hifi-gan taking super long because the recommendation is in steps not epochs so

second: hl1 gman

last one: jeremy clarkson on gran turismo

Oh say less slime

?

My bad it's just UK slang for "oh okay"

are you using it on gpu?

that Gman sample sounds real good

what was the input for this one? sounds great

Could it fix a clip like this?

oh it looks like the voicefixer paper has been published to arxiv

...17 hours ago