# machine-learning
- u
user
11/01/2021, 5:06 PMMy goal is to make confusion matrix for this file - u
user
11/01/2021, 5:06 PMCan anyone help me - m
mega b
11/01/2021, 6:19 PMi think this is normal tacotron2 - m
mega b
11/01/2021, 6:20 PMnot sure - m
mega b
11/01/2021, 6:21 PMnvm it was made 4 years ago - m
mega b
11/01/2021, 6:21 PMthere is no way its tacotron2 - m
mega b
11/01/2021, 6:22 PMbut its whatever we can train our own base model - m
mega b
11/01/2021, 6:23 PMtheir spanish dataset is 11,111 lines long - m
mega b
11/01/2021, 6:24 PMso its literally just LJ-Speech but spanish - m
mega b
11/01/2021, 6:33 PMadd wavs/ to each beginning line of your transcript - e
Edward
11/01/2021, 6:35 PMooh ok - e
Edward
11/01/2021, 6:49 PMyes worked now - e
Edward
11/01/2021, 11:48 PMcan someone with good gpu with much vram help me run openai jukebox. i do not have good gpu and colab is not working. please dm me if you can - a
Alexius08
11/02/2021, 8:29 AMDid I just overfit my model? Versions from earlier epochs pronounce the word "hours" correctly. This one doesn't. - h
hecko
11/02/2021, 5:06 PMyeah it's overfitting, but it's the kind of overfitting that most uberduck voices have anyway - h
hecko
11/02/2021, 5:06 PMhold on lemme get uhhh arpabet notebook - h
hecko
11/02/2021, 5:06 PM - h
hecko
11/02/2021, 5:07 PMtrains on phonemes instead of (or actually as well as) text - h
hecko
11/02/2021, 5:07 PMbut you have to ask zwf to enable arpabet for the model when you submit.it - t
TwoTimmm
11/02/2021, 7:39 PMFun Fact my camille model speaks arpabet nicely even though I didnt even train her on that - t
TwoTimmm
11/02/2021, 7:40 PMThough she kinda pronounces strings as strength, but that happens with the actual word as well - b
BrandonTD
11/02/2021, 8:14 PMcamille from league? - b
BrandonTD
11/02/2021, 8:14 PMi'm training draven currently - t
TwoTimmm
11/02/2021, 9:27 PMOh thats nice! Fun to see more League models - t
TwoTimmm
11/02/2021, 9:27 PMAnd yep - t
TwoTimmm
11/02/2021, 9:27 PMI was also trying to make Riven but it wouldnt let me train her cuz it screamed about the sounds being broken - h
hecko
11/02/2021, 9:35 PMi'm thinking about eventually training a big multispeaker model on every single lol voice - h
hecko
11/02/2021, 9:35 PMincluding every language maybe - h
hecko
11/02/2021, 9:36 PMthough that's an extremely long shot - h
hecko
11/02/2021, 9:36 PMsince transcription is pain and speech recognition doesn't work very well with accents