it sounds like a lady having a stroke while barking idk its weird as hell

hmmm

yeah you'll need to train for longer

just run the training notebook again (and upload dataset and all), it'll pick up from where it left off

ok

@User Thanks for the help tho, will ping if there are further issues

He has came back, check general

i am aware

oh nvm you probably know

yeah

he survived the crash

he just had a coma, that's all

?

some guy was trying to access your website and it was Deleted and thats why i said that

that doesn't make it less weird

I explained to everyone who asked me why took the site down lol

we made a little joke

that's all

Hey, new here, and quick question. I see that uberduck has two modes, one TTS and another that's got the audio reference. I want to know a little more about the second one and was wondering if anyone had the code or notebooks that do the same/similar thing?

the second thing or "Music mode" is talknet.

you can find the notebook at #841437191073955920

is there a version of the notebook that throws out bad audio/transcript pairings because there might be overfitting? at epoch 8 & 17, i was getting a diagonal line at 0.23...now i'm at epoch 68 but the diagonal looks worse at 0.16

don't worry about the line becoming horizontal, that just means the random test sentence was shorter this time (in synthesis it would cut off by then)

ohh that's what it means...

yeah

i thought it was cosine distance

no it's the attention graph

left to right is time, down to up is what character it's looking at

gotcha, thanks

some people call it a spectrogram so at least you're not the wrongest