Sure.

Gimme a sec, I have to upload it on Google Drive.

do you want me to post it in #835647732453605376

yes

it's really weird that it would degenerate so bad after only 64 epochs

maybe it was just bad luck

(there is a degree of randomness in this process)

and the dataset seems really good

Yeah, I guess it's either bad luck or I need to try a different colab.

there is a effect called catastrophic forgetting

that's basically a somewhat sudden collapse of the model

that generative models can suffer from

oh

fun

your best shot in this case would be just to start training from scratch

Alright.

Still, I wish Tacotron could at least tell me "ok you're getting close to overfitting, time to shut it down"

and not be a shot in the dark

@User what notebook do you use for training? (If you're trying again with the same dataset, we should make sure that the random number generator is seeded differently otherwise you'll get the same results)

I used this version of Tacotron.

my voice idea

i'm gonna dm you about this, zwf.

so I don't flood the chat

k

im on a chrome book

oh the owner

I tried put Ronald in the Ai mels but didn't work as it used to.