Yep

fuck me

screenshot your transcript for me jacob

spontaneous combustion. how many clips do you have

alright

time to make the dunk

honestly so far so good

idk what happend to the ending part but it's almost a straight line

epoch 250 on dexter :|

i'm not sure if i could release dunkaccino in this state

he seems...

off

aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

machine learning moment

anybody knows what could cause this collapse to happen?

this inference or training?

training

It's reaching 150 epochs soon, but iit keeps happening here and there, making it seem like that its a bit instable

It is fine. The model trains with batches of audio files. The red square shows where the data is. Above the red square is "_" underscores that the model will ignore, and to the right of the square is empty audio, that the model will again ignore during training. This extra stuff on the sides is just padding so all the inputs have the same length when they enter the GPU.

You can normally ignore everything after the line kinda goes flat

Oh then im glad in that case. Wasnt sure at first so i felt like asking lol

a more promising looking Dexter

a model i let train failed horribly because for some reason the resampler colab screwed up and cut everything to 2 seconds. if i didn't overwrite the original files this wouldn't be a problem but now i have to go through the stupidly long process i went through to get those WAVs again.

how doyou overwrite the original files

Been sitting here wondering how to make my Tacotron2 model read Arpabet input, bc the patch I tried out didnt seemed to work hah

did you download the newest version

Well, I tried with the Playboit Carti Model😦

Wish someone could train it right for me

Oh there was another update rip haha