# machine-learning
- e
Edward
12/05/2021, 7:43 PMcan i train tacotron on amd gpu ? - u
(Dawn) Will Draw Fictional Women
12/06/2021, 9:10 AMheard there was something recently released that would make it theoretically possible - b
BrandonTD
12/06/2021, 2:05 PMWhen transcribing rappers, is it a better practice to use data from a podcast (i.e. them talking) or the rap verses they have themselves? - u
(Dawn) Will Draw Fictional Women
12/06/2021, 2:05 PMdepends on the kinda model you wanna make - u
(Dawn) Will Draw Fictional Women
12/06/2021, 2:06 PMpeople have made models using either - b
BrandonTD
12/06/2021, 2:06 PMGotcha, thanks. - s
SidPlays_144p
12/06/2021, 5:57 PMif you're doing a talking model the podcast is recommended, if you're doing a rapping model the verses are - u
user
12/07/2021, 5:26 PMFuck - u
(Dawn) Will Draw Fictional Women
12/07/2021, 5:27 PM - m
matty1406
12/07/2021, 8:21 PM@User can you delete all of the others as well? - m
Meow
12/08/2021, 11:25 AMhey - does the speed of the person talking matter? if they join 2 words together should I still write out all the words? I'm just nervous about a misalignment - m
Meow
12/08/2021, 11:27 AMe.g. "trying to" when spoken sounds like "tryna" which one to I transcribe - g
Gamma Prime
12/08/2021, 2:01 PMThis is why I'm in no hurry to do Blurr. - g
Gamma Prime
12/08/2021, 2:02 PMHonestly, I don't know if I want to do any more models now. I had high hopes for Beast Wars Megatron, but he doesn't sound very good after training for 250 epochs. - g
Gamma Prime
12/08/2021, 2:03 PMI'm feeling like I need to completely get over Transformers right now. - m
mega b
12/08/2021, 4:11 PMYes the talking speed matters. If you want to hopefully avoid the problem of ruining the model, ignore the clip, and keep the dataset consistent. Otherwise, talknet is a good alternative to tacotron2 that may handle faster speeds. - h
hecko
12/08/2021, 5:10 PMi'd do
in this case but i'm not sure how talknet would react, given it looks all the words up in a dictionarytryna - a
ak
12/08/2021, 6:09 PM - i
IBob012
12/08/2021, 6:09 PMedresson - i
IBob012
12/08/2021, 6:09 PM - a
ak
12/08/2021, 6:10 PMsample - g
Gosmokeless28
12/08/2021, 8:31 PMWhat? - m
matty1406
12/08/2021, 8:31 PMThat was back with the scammers - m
matty1406
12/08/2021, 8:31 PMDoesn’t matter now… - g
Gosmokeless28
12/08/2021, 8:32 PMk - m
mega b
12/08/2021, 10:24 PMAwesome stuff @ak !!!!! - m
Meow
12/08/2021, 10:27 PMAhh 50% of my samples are all really fast talking - h
hecko
12/08/2021, 10:35 PMthen make a separate model with just the really fast talking - h
hecko
12/08/2021, 10:35 PMit should work, haven't tested it though - m
Meow
12/08/2021, 11:30 PMoh yeah smart