# machine-learning
- u
(Dawn) Will Draw Fictional Women
05/04/2022, 6:38 PMrevise the dataset - m
myposynator
05/04/2022, 6:41 PMHelp, I don't know what to do - h
hecko
05/04/2022, 6:43 PMdo you want to make a model and don't know where to start? - m
myposynator
05/04/2022, 6:45 PMWell, I think so. - m
myposynator
05/04/2022, 6:46 PMWell, I think I did everything right, but the graphics are horrible - i
imphy
05/04/2022, 6:50 PMIs there somewhere that will explain why this is bad? How do I read these graphs to determine if it's good, bad, or great? - m
myposynator
05/04/2022, 6:55 PMThis is one of the doubts I have. I've already tried to make two tacotron models but none look like the "good" chart, or like the reference chart. - h
hecko
05/04/2022, 6:57 PMuhhh you might be training it for too long - h
hecko
05/04/2022, 6:57 PMwhat's the last point where the graphs looked decent - m
matty1406
05/04/2022, 6:58 PM200 epochs isn't that long in my opinion, its only at 0.10 validation loss - u
(Dawn) Will Draw Fictional Women
05/04/2022, 7:08 PMstaircase broken - u
(Dawn) Will Draw Fictional Women
05/04/2022, 7:08 PMthis is just a goopy mess - d
dzlandis
05/04/2022, 7:14 PMIs Talknet only for music singing stuff? Or is it a better version of Talcotron? - u
(Dawn) Will Draw Fictional Women
05/04/2022, 7:18 PMits only for reference audio on uberduck - u
(Dawn) Will Draw Fictional Women
05/04/2022, 7:18 PMit can speak normally off of just text - u
(Dawn) Will Draw Fictional Women
05/04/2022, 7:19 PMbut its also very strict with how it likes its datasets for it to work well - u
(Dawn) Will Draw Fictional Women
05/04/2022, 7:39 PMaaaaagh why wont you fiiix - u
{K EY1} (Kei)
05/04/2022, 8:35 PMI removed some instrumentals from some songs to make a singing model for talknet and nnsvs, but there's still some sounds left on some samples Would a noise removing program (rather than an instrumental remover) work for that? - w
Waltman13 (semi-active)
05/04/2022, 8:38 PMis this star? - h
hecko
05/04/2022, 8:40 PMnoise removal usually targets consistent noise so no not really if you have an instrumental i recommend using that to phase-cancel it out and then run vocal remover - u
{K EY1} (Kei)
05/04/2022, 8:41 PMI don't think there are official instrumentals of the songs in question - u
{K EY1} (Kei)
05/04/2022, 8:42 PMThe noise is sorta consistent? Like i mean using something like audo.ai or whatever - h
hecko
05/04/2022, 8:44 PMah - h
hecko
05/04/2022, 8:44 PMaudo.ai could work potentially - w
Waltman13 (semi-active)
05/04/2022, 8:45 PM🙃 - w
Waltman13 (semi-active)
05/04/2022, 8:46 PMmaybe I altered her dataset a lil too much - u
(Dawn) Will Draw Fictional Women
05/04/2022, 8:48 PMcheck the last version i sent in #835647732453605376 for mistakes (i am sure they will be obvious and i will be stupid) - w
Waltman13 (semi-active)
05/04/2022, 8:50 PMalright. I'll have to check later tho as I'm going to the store in a bit. - r
Reclezon
05/04/2022, 9:15 PM