im sure some of them are, but i couldn't recommend making a purchase with no recourse

stupid google

but thanks for the help anyways

ill have to keep asking around

(or just use colab)

but internet sucks and uploading files slow

colab is nice - the files are a bummer - at least its faster in the synthesis notebook

i will start upload now for training n stuff

bye bye

My guess is you're using slightly different libraries, maybe you're missing the vocoder or a dictionary or something

Is everything 1:1 identical?

Sounds like you're not configuring your hparams/any optimizations or loss criterion to match the collab

Just copy paste all the collab code and change out the parts that assume you're using a website vs. Local.

Also dont skip converting the wavs into npys

yeah

that shows you the graph to see how well your voice is being trained

very important

not sure what you mean that it's extremely good at making instrumentals, just not so much the vocals

umm

I don't think this'll turn out good

oh wait nevermind

SOMEONES MAKKING EDDS WORLD?

I completly forgot about that cartoon

thats a rlly good start! keep it going i cant wait to see what it sounds like

There are already a lot of those guys on the site lol

this is about 250 audio tracks and im still getting more. :O this sounds very cool so far.

i think i'll give up on trying to make the cg5 dataset all the acapellas i made so far have at least one of: dips on every kick drum beat which the ai would replicate (the main issue), a reverby clip before the vocals so i can't cleanly cut the lines, or (in one case so far) high-pitched instruments that didn't get clipped out and for some reason i can't bring myself to just try anyway

that being said if anyone here knows of a singer who regularly publishes official acapellas i'd be glad to make a dataset and model of them

...hmmm i could use the main-vocal-only-but-kinda-bad model and mix it with the all-vocals-but-good one

but even then that only solves one song, 30 seconds of data at most aaa