# machine-learning
- u
{K EY1} (Kei)
08/12/2022, 1:39 AMIf you trained it on the talqu base model, that would be because the talqu base model is very limited in data and only has one speaker - u
{K EY1} (Kei)
08/12/2022, 1:40 AMI always train with the neutalk phonemes and base model now - p
PUMPKINEATER
08/14/2022, 6:20 AMDid you see it? - j
Justin
08/14/2022, 6:43 AMNuke's Top 5 moment - w
WLGT-TV Channel 10 New York
08/14/2022, 1:17 PM@everyone how do i submit a voice like this? - c
cadeotsg
08/14/2022, 7:12 PMdid you expect to ping everyone in this 2000 player server without any roles - u
(".Lukus_bt.")
08/14/2022, 10:10 PMtesting my jon from eddsworld model - c
Cris140
08/15/2022, 10:35 PMI remember trying to train a multispeaker model with around 5~6 voices, and they got scrambled, I think I saw other people having the same problem, is that fixed? - u
(Dawn) Will Draw Fictional Women
08/15/2022, 10:35 PMnope - u
(Dawn) Will Draw Fictional Women
08/15/2022, 10:35 PMwell - u
(Dawn) Will Draw Fictional Women
08/15/2022, 10:35 PMmaybe - u
(Dawn) Will Draw Fictional Women
08/15/2022, 10:36 PMbut if they are know that its order by python alphabetical and not standard numerical - h
hecko
08/15/2022, 10:56 PMthe scramble only happens at 11 voices and up though - u
(Dawn) Will Draw Fictional Women
08/15/2022, 11:00 PMyea because it does 1,10,2 instead of 1,2,[…]10 - c
Couch
08/15/2022, 11:08 PMgo go gadget pip install natsort (natural sort (will sort numerically (or actually just cast the speaker ids to int instead of str))) - c
Cris140
08/15/2022, 11:42 PMStrange, I didn't had that much speakers - c
Cris140
08/16/2022, 6:22 PM - c
Cris140
08/16/2022, 6:22 PMDid anyone happened to test this? It looks faster on CPU than hifigan - l
Lexi (delulu posts on the daily)
08/16/2022, 11:29 PMbut is it better? - c
Cris140
08/16/2022, 11:52 PMThat's what I wanted to know - x
Xudong.zhong
08/17/2022, 5:30 AMsed - b
Booce
08/17/2022, 7:15 AMAre the graphs meant to look this weird sometimes? - u
{K EY1} (Kei)
08/17/2022, 2:26 PMYeah, usually that just means it's showing a shorter wav - c
Couch
08/17/2022, 2:37 PMthe recommended minimum length is 1.2~2 seconds max around 12 - b
Booce
08/17/2022, 2:39 PMI knew the maximum length, didnt know the minimum length though. I know for a fact some of the wavs are about 1 second long, would it be worth it to take those out? - c
Couch
08/17/2022, 2:39 PMyou can merge several of them together in your favorite audio editor that way you dont lose data - b
Booce
08/17/2022, 2:39 PMAlright - j
Justin
08/17/2022, 3:08 PMyou can fuse shorter clips together and make then longer - j
Justin
08/17/2022, 3:08 PMkeep it under 6-10 secs - u
{K EY1} (Kei)
08/17/2022, 10:05 PMI don't really pay attention to minimum length and i haven't had problems