sample?

un momento

aint no way

you trained 1000

for chowder

The dataset totaled 2 minutes in duration, so I had to train the model for more epochs than normal

dawg

how long did it take

i've trained <1 minute models on like 300-500 you kinda wasted some time 😭

No I didn't bruh, I trained for as long as I could _right until_ the model started to become overfit.

🤷‍♂️🤷‍♂️🤷‍♂️

Besides, the model trained pretty damn fast, it felt like the training went by 1 epoch per second

oh yea that happens with smaller datasets yeah

Less than an hour, I think

curious tho why only 2 minutes of audio

Ask

i make mine with that amount of data at 400 and it sounds fine this has gotta be overtrained as SHIT

Trust me, I monitored the graph, and I saved the checkpoint that was created right before the model started to become overfit.

The loss/total graph

Wheezer says he wasn't able to find more storyboard video versions of episodes

uvr5 to the rescue lole

dawg i have message perms now

Nice BTW, what Cassidy means is that you can use a speech isolator like UVR 5 to remove background noise from voice clips

ik

i just dont have the patience to do that

Plus, it's not always recommended

lord forgive me

Thoughts?

he let the intrusive thoughts win

It sucks that these retards' unconstructive feedback was archived. 👎 I tried telling them that 1,000 was how many epochs the model had to be trained for, but they were saying retarded shit like "you wasted time bro," "this shit's gotta be overtrained as hell," "I don't need to train my RVC models for that many epochs for them to generate good-sounding results" That feedback genuinely gets me angry, so fuck those dicks.

"good morning everybody! whoa, what happened in here?"

This model dry as sukuna