well the training/model broke, gotta restart it...

stuck at 267mb

is there like a general upper limit to the amount of speech you should have per line?

I've chunked like 20 mins of audio up into ~200 files

oh of course it's in the faq, thanks

Up to 12 seconds is the recommended, but I can confirm that a wav file of 27 seconds also works. (Though it's more memory intensive)

gotcha, yeah. Looks like I've got a good variety but one goes up to 32s oops

I've been trying to get this model to train effectively for the past few days on and off, it's all the small things haha

pain

another question while I'm being a bother, would extra empty audio at the end of clips be likely to cause problems?

Unfortunately, the uberduck.ai bot has not gotten verified by Discord yet even though it should have been by now. It has been taking a long time for Discord to verify the bot for some reason.

nice try dyno

lol

Sometimes. If it is too drastic, lets say more than 1 second of silence, it may cause problems.

Gotcha, thanks. Trying to see if I can trim what I've got with ffmpeg to see if it helps

yes, probably

when i make a sentence it just comes out as 34 seconds and after finishing the sentence it just makes weird af noises

I had that issue with the BJ model.

ah

finally cleaned up my data so all my clips are under 14 seconds and its turning epochs much faster

it still takes about 15 seconds an epoch and will take around 55 hours to train gonna just let it run all night and see what happends

at least its not over two weeks anymore

read it https://pieksend.github.io/posts/2021-05-09-tacotron-faq/

yes thank you you sent it before

but you didn't read it

it said they should be 2 seconds and under 12 seconds but i did do it