speakers may say the numbers differently than what infer converts it to

hmm what if i were to mark word emphasis in the training data like google did with their model @

Tacotron learns stress and intonation.

but idk if our version even looks at capitalization (and if it does it's likely completely separate letters which is hella jank) so maybe like

The buses aren't the #problem, they actually provide a #solution.

in this episode of trying to salvage a model but actually making it worse:

yeah nvm i'm training from scratch, it's probably gonna take less time and be better

checked the tacotron2 source code and it seems that by default

#

isn't supported, maybe parentheses or semicolons or such exclamation marks would make sense but also would collide with actual exclamation marks

a promisingly looking epoch from chef fritz

Fritz needs some retraining

Um, I got some bad news actually, it's not there anymore.

señor pi died in a car crasj )he left the serv er =

Meet the Scout but the Scout dies.

LOL thats really good

thanks for sharing haha

Can anybody make a lil uzi voice?

You can request it doing the voice suggestion command

Oh ok

can you post me a link to this generator?

Added some more Fritz lines and this graph looks a lot more promising than the other

@User hey, how is the King Julien TalkNet voice coming along? I haven't heard anything about it in a while from you or Carl Da KWASON, so I thought I'd ask you about it.

um

ill work on it

cause i have other things to work on

soon tm

oh ok.

vocalremover5 abruptly cuts vocals for me!

do you have

postprocess

enabled? if so then that's what's doing it, disable it

also see this for a potentially better model

Okay, so I was about to train the Cranky Kong dataset, and-*HOLY SHITTING SHIT!*

"DKC bonus fail theme plays"

what is your batch size?