# machine-learning
- n
Needlemouse
09/26/2021, 4:35 PMlike how you added character specific voice clips - s
Swiper
09/26/2021, 5:36 PMyeah I just tried out the batch converter had to modify the code since my files were already 22050 and I just needed them to be mono and then it cut all the wavs down to 2-3 seconds fml gonna just have to use audacity - u
user
09/26/2021, 6:45 PMalright I need to learn how to do this - u
(Dawn) Will Draw Fictional Women
09/26/2021, 6:46 PMmodel name: well duh - u
(Dawn) Will Draw Fictional Women
09/26/2021, 6:46 PMtraining file: if you named it list and put it in the right place you can just leave that - u
(Dawn) Will Draw Fictional Women
09/26/2021, 6:46 PMbatch size: change it depending on the gpu you got - u
(Dawn) Will Draw Fictional Women
09/26/2021, 6:47 PMepochs: 250 is good enough - u
(Dawn) Will Draw Fictional Women
09/26/2021, 6:47 PMoutput directory: where save - s
Swiper
09/26/2021, 7:06 PMif you trained a model using tacotron 2 can you still use the talknet synthesis? - u
(Dawn) Will Draw Fictional Women
09/26/2021, 7:06 PMwh - u
(Dawn) Will Draw Fictional Women
09/26/2021, 7:06 PMno - u
(Dawn) Will Draw Fictional Women
09/26/2021, 7:07 PMwhen could you ever do that - s
Swiper
09/26/2021, 7:11 PMfirst time making something was just wondering - s
Swiper
09/26/2021, 7:11 PMalso how do you get a better gpu every time I factory runtime reset I always get Tesla K80 I can't seem to get anything else - h
hecko
09/26/2021, 7:14 PMyou don't - h
hecko
09/26/2021, 7:14 PMeither google ran out of them or they intentionally downgraded the free tier or both - s
Swiper
09/26/2021, 7:15 PMcan I still train using k80 it'll just take longer or do I gotta pay for better gpu - h
hecko
09/26/2021, 7:15 PMyeah it works - h
hecko
09/26/2021, 7:15 PMtakes 2.5x as long i think - u
user
09/26/2021, 9:25 PMi tried submitting my karateka vocal dataset in both #842056330835656745 and here, but it still is failing to interact with me - h
haru0l
09/27/2021, 12:55 AMhuh the batch converter always worked fine for me - h
haru0l
09/27/2021, 12:55 AMthough i export to another folder since i want to save the origs - h
haru0l
09/27/2021, 12:56 AMweird that ffmpeg cuts audio though, you'd think people would notice this - c
cengi
09/27/2021, 8:43 AM. - c
cengi
09/27/2021, 8:44 AM. - h
hecko
09/27/2021, 8:48 AMit means the model didn't press the "ok i'm done" button hard enough, so the notebook kept asking it to continue reading, but since there's nothing left to read it ends up mumbling - h
hecko
09/27/2021, 8:49 AMmost likely cause is you forgot a dot at the end second most likely is your model was trained on a low amount of clips so it'll do that sometimes, try putting two dots to help it along - c
cengi
09/27/2021, 8:49 AMdoes it do that in the notebook you provided ? This happens often on the normal tacotron training - c
cengi
09/27/2021, 8:49 AMOh. - u
user
09/27/2021, 11:28 PMAAAH