Works fine now

😮

do you just replace the models in the zip?

interesting, does it grant better sounding models?

not sure about that coz I used the old models so that the training begins

I will take anything that works lol I was trying to train last night and I was like, this hasn't happened before lol

This is just a temporary workaround

until they update the code (SortAnon)

@User thanks for the work around I really needed this

np

Looks good so far

training filename2 and funny enough, graphs look better than phone guy...

lol

hows the progress on voicefixer phoneguy matty?

haven't tried it yet, for some reason, it feels like it takes years. i tried without voice fixer first and you were right, it sucks. i will probably try phone guy with voice fixer after filename2.

ok nvm, graphs lied...

How does it work?

How do you do this?

how do i stop google from kicking me out while im doing my training

you don't™

best you can do is keep the tab open, do the kapshkas every few hours, and once it kicks you out anyway just re-run it with the same filename and folder and it'll resume training

where can I read about controllable talknet? I have read talknet research paper.

i don't think it's officially documented anywhere, all i've seen is a 4chan post where sortanon said he made the pitch input into 2 inputs (basically the x and y coordinates of a point on a pitch circle that loops every octave)

Oh, thanks @User , So its basically take place at inference level, no change in training process? Rest i'll have to study about the octave change.

nono it changes the training too

oh, I'll check that, thanks again!

thanks guys

throwing this here if anyone makes heads out of it

does this mean mine is already doomed for failure lol