# machine-learning
- u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:06 AMbut now that you mention it daniel sorely needs it - p
PUMPKINEATER
03/27/2022, 6:06 AMYeah. - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:06 AMdee bee kaw and oliver are pretty damn good tho - p
PUMPKINEATER
03/27/2022, 6:06 AMMhm. - p
PUMPKINEATER
03/27/2022, 6:06 AMGod damn man. - p
PUMPKINEATER
03/27/2022, 6:07 AMUsing my new Audio2Dataset colab - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:07 AMi was thinkin specifically anji mito, may, and sakura - p
PUMPKINEATER
03/27/2022, 6:07 AMI am getting 5K for Obama's audiobook - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:07 AMsakura specifically is REALLY promising but definitely bogged down by the clips where she yells her move names - u
{K EY1} (Kei)
03/27/2022, 6:07 AM1. Morty smith 2. Eminem freestyle 3. Meowth - p
PUMPKINEATER
03/27/2022, 6:07 AMNot Uberduck's - p
PUMPKINEATER
03/27/2022, 6:07 AMthe one I showed - u
{K EY1} (Kei)
03/27/2022, 6:08 AMI am so sorry - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:08 AMi might ask blatant to retrain may without her yelling clips but idk how busy he is - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:08 AMykw @User how busy are yas - u
{K EY1} (Kei)
03/27/2022, 6:08 AMI completely missed the part where you were showing something I feel like an idiot now. - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:11 AMdamn - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:11 AMnoisepuppet tacotron bouta keep me up to 3 am - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:51 AMoh fuck you know what also needs this treatment but i really keep putting off - u
(Dawn) Will Draw Fictional Women
03/27/2022, 6:51 AMthe left 4 dead talknets - b
big black (blatant) balls
03/27/2022, 1:16 PMIdk I can probably do it - h
Heavyser
03/27/2022, 3:53 PMCouldn't find a media channel, so I'll post this here - h
hecko
03/27/2022, 4:17 PM#841461499946860576 - m
myposynator
03/27/2022, 10:21 PMHi guys, how are you? Sorry for my english, I'm using google translator to communicate. Anyway, I have some problems and doubts about talknet. This week I made a dataset with more than 2000 samples, and here comes the problem. I use colab free i.e. gpu k80. During training, I had a problem at step 9. Apparently CUDA ran out of memory, so I lowered the batch size to 20 and it didn't help, then lowered it to 150 epochs. I would like to know if the model will have a good quality if I train it with these parameters. If not, is there a way to eliminate the problem of CUDA running out of memory? (I've trained the same model but with 1000 samples, 200 epochs and 32 batch sizes, and the sound didn't please me so much) - m
myposynator
03/27/2022, 10:21 PMSorry for the huge text. - u
(Dawn) Will Draw Fictional Women
03/27/2022, 10:32 PMjust continue to lower the batch size - u
(Dawn) Will Draw Fictional Women
03/27/2022, 10:32 PMdont lower the epoch number - u
(Dawn) Will Draw Fictional Women
03/27/2022, 10:32 PMthat just means itll train for less time thus giving a lower quality result - m
myposynator
03/28/2022, 12:27 AMOkay, thanks! - r
Reclezon
03/28/2022, 1:51 AM