# machine-learning
- h
hecko
09/10/2021, 3:51 PMbecause bigger batch means it only changes the network in ways that make the network better at a majority of the clips in a batch, instead of fumbling around trying to satisfy each individually - h
hecko
09/10/2021, 3:52 PManother way is to yoink a similar voice with good quality and train it a bit more on your dataset - h
haru0l
09/10/2021, 3:52 PMyou can try using another trained model that sound similar to your model too instead of using the default one - u
(Dawn) Will Draw Fictional Women
09/10/2021, 3:59 PMwould the same method of training with larger batches improve a model with a dataset of only 94? - u
(Dawn) Will Draw Fictional Women
09/10/2021, 3:59 PMor is it just a dataset issue at that point - m
matty1406
09/10/2021, 3:59 PM94 should be an alright amount, anything above 60 audio files i say is doable. - u
(Dawn) Will Draw Fictional Women
09/10/2021, 4:00 PMthe thing is quite a few of the lines are retakes of the same line - u
(Dawn) Will Draw Fictional Women
09/10/2021, 4:00 PMso i think that might be skewing a bias - u
(Dawn) Will Draw Fictional Women
09/10/2021, 4:01 PMi.e. the voice i'm working on is patricia wagon and she has five different title reading lines - u
(Dawn) Will Draw Fictional Women
09/10/2021, 4:01 PMso like five lines that are just "mighty switch force academy" for example - u
(Dawn) Will Draw Fictional Women
09/10/2021, 4:06 PMshould i remove them or just leave them in and try training with larger batches anyway - u
user
09/10/2021, 7:33 PMhow do make a voice for uberduck - h
haru0l
09/10/2021, 7:51 PM#841422801965416538 - p
PUMPKINEATER
09/10/2021, 8:36 PMI thought the default voice is LJSpeech - z
Zachary(retired f/ voice clonin)
09/10/2021, 8:37 PMChills is the Link2Model LJSpeech - h
hecko
09/10/2021, 9:04 PMooh this is pretty impressive https://www.lovo.ai/ - h
hecko
09/10/2021, 9:04 PMvery impressive if there's no reference audio, i can't tell whether there is - h
Hugh G. Rection
09/10/2021, 9:10 PMI did also train the CDA agent from the Monsters, Inc. Scream Arena game as well, but... I tried making him say "We got a number one victory royale," but this happens: - u
user
09/10/2021, 9:12 PMThat happens, I believe it's because it tries to generate more than it should - e
Echoes
09/10/2021, 10:16 PMis there a way to use the training notebook on jupyter lab? - w
Waltman13 (semi-active)
09/10/2021, 10:21 PMI thought he said "We get an.. ERROR!!" - u
(Dawn) Will Draw Fictional Women
09/10/2021, 10:21 PMhows this soundin to yall - u
user
09/10/2021, 10:21 PMCool! I like Red Hot Chili Peppers! - u
(Dawn) Will Draw Fictional Women
09/10/2021, 10:22 PMaight, i'll run another test prompt at it and try submitting the model - u
user
09/10/2021, 10:22 PMOk - u
user
09/10/2021, 10:38 PMwhat does it mean when a voice model is doing something like this? - c
Couch
09/10/2021, 10:38 PMend of clip - u
user
09/10/2021, 10:38 PMoh k - u
user
09/11/2021, 12:48 AMMrwhosetheboss training going smoothly