# machine-learning
- j
J CrashBandicoot F
02/12/2022, 2:39 AMNo idea what this means.. and I never had that error before, that was on step 4 - u
(Dawn) Will Draw Fictional Women
02/12/2022, 2:42 AMall i can recommend is checking your dataset - p
Pathos
02/12/2022, 2:08 PMIts learning!! - h
haru0l
02/12/2022, 3:31 PMi honestly have no idea anymore how to combine languages - h
haru0l
02/12/2022, 3:32 PMwould not using phonemes on tacotron work - h
haru0l
02/12/2022, 3:32 PMlike English text and then hiragana - e
Emoto
02/12/2022, 8:14 PMNo idea, it is possible to do cross langual synthesis tho, Google has proven it before, but I think there's no public SRC out for that yet - h
hecko
02/12/2022, 11:43 PMthat's how i'd do it yeah - h
haru0l
02/13/2022, 9:03 AMit it normal for gdown not to work sometimes - h
haru0l
02/13/2022, 9:03 AMsometimes itll complain that the file isnt accessible due to permissions but rerunning it again dls the file fine - s
SpamZow (WoomyPearl)
02/13/2022, 2:51 PM - s
SpamZow (WoomyPearl)
02/13/2022, 2:51 PMMy Doug Dimmadome dataset doesn’t train properly. - s
SpamZow (WoomyPearl)
02/13/2022, 2:52 PMIt always turns into him stroking no matter what settings I change. - s
SpamZow (WoomyPearl)
02/13/2022, 2:52 PMCan anyone check and see what went wrong? - j
joao0213
02/13/2022, 5:48 PMLimao any tool suggestions for speaker diarization? - p
prism
02/14/2022, 8:05 PMI found descript to be really easy to use for speaker diarilzation. That and the studio sound audio cleaner makes it worth it. Been trying to figure out a way to use it for cuts - r
roihttps
02/14/2022, 8:36 PMhelp please - r
roihttps
02/14/2022, 8:50 PMand i dont have any "<" in the transcription - p
Pathos
02/14/2022, 9:47 PMlooks like it couldn't load the pickle - j
J CrashBandicoot F
02/14/2022, 10:05 PMTried to warm start from chills with a 17 wav dataset, and it turned out like this - j
J CrashBandicoot F
02/14/2022, 10:07 PMIs it alright? Should I train it more or did I trained it so much? - h
hecko
02/15/2022, 2:06 PM0.06 loss looks like too much yeah - h
hecko
02/15/2022, 2:06 PMbut who knows maybe it's good - h
hecko
02/15/2022, 2:07 PMlast time i thought i overtrained a small-dataset model the person who requested it ended up preferring the overtrained version over a backup - j
J CrashBandicoot F
02/15/2022, 3:39 PMIt didn't sounded good, it was stroking a lot and wasn't saying most of the words normally, and also very robotic voice, probably it's the voice I was trying to train I'm not sure.. - j
J CrashBandicoot F
02/15/2022, 3:41 PMAnd I even used this step 3 replacement made by Dawn which auto calculates the learning rate based on the batch size you place - u
(Dawn) Will Draw Fictional Women
02/15/2022, 3:47 PMsometimes a dataset can have completely clean audio and a completely accurate arpabet dataset and still just shit itself