# machine-learning
- m
mega b
02/16/2022, 12:37 AMbut but - m
mega b
02/16/2022, 12:37 AMnot sure about dat - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:38 AMshould it be hybrid like ive been doing before with normal taco - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:38 AMi.e. - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:38 AMCopy codewavs/12.wav|What can I say? The skill is undeniable. S'only one ace me. wavs/12.wav|{HH W AH1 T} {K AE1 N} {AY1} {S EY1}? {DH AH0} {S K IH1 L} {IH0 Z} {AH2 N D IH0 N AY1 AH0 B AH0 L}. {S OW1 N L IY0} {HH W AH1 N} {EY1 S} {M IY1}. - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:39 AMwill torchmoji apply to both transcripts or only the lj one - m
mega b
02/16/2022, 12:40 AMthe original warm_start model we train with on the colab is already hybrid ๐ - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:40 AMoh! - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:40 AMso this is ideal - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:40 AM? - m
mega b
02/16/2022, 12:40 AMyep - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:40 AMfuck! im already in business! - m
mega b
02/16/2022, 12:40 AMCHA CHING - m
mega b
02/16/2022, 12:40 AM๐ธ - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:41 AMhell yeah ill start figuring the actual notebook out now - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:41 AMafter i set up my multispeaker dataset - u
(Dawn) Will Draw Fictional Women
02/16/2022, 12:42 AM(btw pray tell how do i format multispeaker again ๐ณ ) - s
samkoelle
02/16/2022, 3:52 AMyou have to add |0 after all the wavs for the first speaker, |1 for all the wavs for the second speaker, |2 for the third and so forth - s
samkoelle
02/16/2022, 3:53 AMworking on making this automatic so you can just have a folder of folders containing (wavs/, list.txt) for each speaker - h
hecko
02/16/2022, 9:22 AMi already made a script for that ๐๐๐ ๐ - h
haru0l
02/16/2022, 9:52 AMis there any tips yet for training a model with a really small dataset - h
haru0l
02/16/2022, 9:53 AMother than "get more wavs" - h
hecko
02/16/2022, 11:00 AMuhh train a talknet model i think - h
hecko
02/16/2022, 11:01 AMand in preparation for that make pitch-shifted copies of your wavยทs so it can be mildly decent at singing - h
hecko
02/16/2022, 11:02 AMor: wait until multispeaker models are supported by uberduck, then train one on your tiny dataset along with some similar voices - h
haru0l
02/16/2022, 1:25 PMuhhhh - h
haru0l
02/16/2022, 1:25 PMim on the inference notebook btw - h
haru0l
02/16/2022, 1:28 PMoh uploading the config made it work - h
haru0l
02/16/2022, 1:28 PMi wonder whats wrong