# machine-learning
- c
Couch
01/02/2022, 11:11 PMi scroll down to the bottom of the list and i see synthesis results - m
matty1406
01/02/2022, 11:47 PMmade a voice... didn't sound as great as I expected but still decentish - i
IBob012
01/02/2022, 11:48 PMi didnt even need that second audio file to guess who it was lol - i
IBob012
01/02/2022, 11:48 PMthe vocoder hates his voice its better to train a hifigan model just for him - u
(Dawn) Will Draw Fictional Women
01/02/2022, 11:55 PMthat might just be lj - u
(Dawn) Will Draw Fictional Women
01/02/2022, 11:55 PMuber’s alex trebek hifi might suit him better - h
HorseFolder
01/03/2022, 2:23 PMsome good stonks with Wendy Testaburger - g
Gushiavo_N2
01/03/2022, 2:45 PMhow many epochs and batch size i does? - w
Waltman13 (semi-active)
01/03/2022, 2:47 PMI think the problem with Pinocchio is that the quality is really inconsistent on the clips you gave it. Like there's ones that sound normal (albeit cut off a bit too early in some), ones that sound like they're ran through voice fixer, the talknet samples possibly, and I think this one clip here which sounds like it came from Shrek 4D or somethin. Cause there's a lot of echo in it. - c
Couch
01/03/2022, 2:48 PMrun the original dataset through rx 8 and remove inference results - c
Couch
01/03/2022, 2:48 PMbecause i dont think ever in my life have i seen anyone use inference results in a dataset before - m
matty1406
01/03/2022, 4:41 PMi couldnt finish the hifi gan cause it errored for some reason, but I tested it at 2000 steps and I don't hear a difference... will it just be better to just send it without the hifi gan and let it use uberduck's? - u
(Dawn) Will Draw Fictional Women
01/03/2022, 4:44 PM2000 is pretty low so yeah it probably wont do anything - u
(Dawn) Will Draw Fictional Women
01/03/2022, 4:45 PMsee if you can resume training on like an alt or something l - m
matty1406
01/03/2022, 4:46 PMthe sound is definitely by the vocoder, right? i don't want to spend ages and then the quality not be good... - u
(Dawn) Will Draw Fictional Women
01/03/2022, 4:46 PMit most likely is - m
matty1406
01/03/2022, 4:48 PMwow googles being genorous today, still got connected - m
matty1406
01/03/2022, 6:15 PMok now google doesn't like me again - m
matty1406
01/03/2022, 6:15 PMi got kicked out - m
matty1406
01/03/2022, 6:16 PMim gonna use an alt, but how do you get past that new google drive mounting? - u
(Dawn) Will Draw Fictional Women
01/03/2022, 6:21 PMhang on i got that figured out - u
(Dawn) Will Draw Fictional Women
01/03/2022, 6:21 PMlemme find the solution again - c
Cris140
01/03/2022, 10:22 PMdrive.mount('/content/drive', force_remount=True, use_metadata_server=False) - u
user
01/03/2022, 10:28 PMhi is this broken? https://colab.research.google.com/drive/14X73UiywnoL9VS30iPDcX4WXxwZWv2e2 - i
IBob012
01/03/2022, 11:19 PMthats normal, that happens when you have one or more audio files that pass 12 seconds next time you try training a model lower the batch size - m
matty1406
01/04/2022, 3:53 PMokay i trained hifi gan for 6000 steps now... - m
matty1406
01/04/2022, 3:53 PMim testing it and i cant hear a difference tho.... - m
matty1406
01/04/2022, 3:57 PMdid i do the hifi gan wrong or what? it still has noise...