its in the training notebook

awesome because all of my fnaf audio files are 44khz 💀

it doesnt cut the audio files in half right

idk

but it downpitches it

It's fixed

fuck glad i caught it

im gonnna start over now

when i make sure its in 22khz, i just chuck everything in audacity and just set project rate to 22khz and export multiple, super easy!

sometimes audacity does this weird thing where it just

doesnt

export in 22khz

after i set it to

How much should i let the val loss reach in the uberduck pipeline training notebook?

0.10 prolly, not sure watch the loss by the tensorboard tho

You mean the graphs from the scalars section?

about 60k steps into my Any2Any voice changer. Uses PPG, f0 and some other features and output is 36 khz. Latency will only be 80 milliseconds, woooo.

👀

super cool

for multilingual voice style transferring

this happens when i try to test out any of my models. can anyone help?

which file is this from?

is it normal for it to stop here 🏃‍♂️

ah its because gdown broke again

there ended up yeeting gdown with curl LMAO

a downside of this is that it only works once per session (if anyone was interested here's the fix https://github.com/wkentaro/gdown/issues/146#issuecomment-1041572016)

the end code should be like

!curl -L -s -o 'TTSmodel.pt' 'https://drive.google.com/u/0/uc?id={model_id}&export=download&confirm=t'

by replacing gdown

a bit of code to get around gdown not working on talknet

%cd models
!curl -L -o vqgan32_universal_57000.ckpt 'https://drive.google.com/u/0/uc?id=1wlilvBtlBiAUEqqdqE0AEqo-UKx2X_cL&export=download&confirm=t'
!curl -L -o hifirec 'https://drive.google.com/u/0/uc?id=12gRIdg65xWiSScvFUFPT5JoPRsijQN90&export=download&confirm=t'
%cd ..

(fixes the reduce metallic noise (slow) option)

model_id = "1D4rOj6fiaV4D3TlKuJKhhotb54bC7Ppc" #@param {type:"string"}
!curl -L -o model.zip 'https://drive.google.com/u/0/uc?id={model_id}&export=download&confirm=t'
!unzip model.zip -d /content/models/{model_id}

(dls the models)