# machine-learning
- h
hecko
06/29/2022, 11:11 AM
is the name of the default zip compression algorithmdeflate - h
hecko
06/29/2022, 11:12 AMhere it tells you how much each file is compressed because whoever made the notebook didn't tell it to shut up - p
Pathos
06/29/2022, 11:50 PMSome progress on the voice changer I am working on. Still doesn't convert singing or whispering, buuut it does run in realtime on consumer GPUs - u
{K EY1} (Kei)
06/29/2022, 11:51 PMNice! - z
zwf
06/29/2022, 11:52 PMit sounds really good 😀 - u
(Dawn) Will Draw Fictional Women
06/29/2022, 11:52 PMwhats required to get a new voice onto it? - p
Pathos
06/29/2022, 11:52 PM3 minutes of a target voice and about an hour of training - u
(Dawn) Will Draw Fictional Women
06/29/2022, 11:52 PMooo damn - c
Couch
06/29/2022, 11:53 PMwhat if i have more than three (3) minutes of audio - p
Pathos
06/29/2022, 11:53 PMMore audio did seem to improve the quality, but the biggest difference was about 3 minutes. - u
(Dawn) Will Draw Fictional Women
06/29/2022, 11:55 PMhmmm - u
(Dawn) Will Draw Fictional Women
06/29/2022, 11:55 PMcould i ask to hear how a dataset i have would sound? - u
(Dawn) Will Draw Fictional Women
06/29/2022, 11:55 PMif the training isnt too much of a bother - u
(Dawn) Will Draw Fictional Women
06/29/2022, 11:55 PMooh wait how does it do with dialects and accents - u
(Dawn) Will Draw Fictional Women
06/29/2022, 11:56 PMlets say i feed it like uhhh - u
(Dawn) Will Draw Fictional Women
06/29/2022, 11:56 PM30 minutes of eddache who's a brit - p
Pathos
06/30/2022, 12:01 AMSome amount of accent is converted - u
(Dawn) Will Draw Fictional Women
06/30/2022, 12:03 AMmmmm - i
IBob012
06/30/2022, 12:08 AMreally awesome, i wonder how it would sound if i use for example that moistcritikal voice and then speak in portuguese - u
{K EY1} (Kei)
06/30/2022, 12:09 AMOoh yeah if it had crosslingual that would be super cool - u
{K EY1} (Kei)
06/30/2022, 12:09 AMAnd do you need to train on your own voice too? Or just the target voice Also can it work on cpu - u
(Dawn) Will Draw Fictional Women
06/30/2022, 12:10 AMwould the amount of data affect how much accent is carried? - u
(Dawn) Will Draw Fictional Women
06/30/2022, 12:10 AMlike lets say 6 minutes of noel gallagher versus the aforementioned 30 minutes of eddache - u
(Dawn) Will Draw Fictional Women
06/30/2022, 12:10 AMwhich would sound more british - c
Cris140
06/30/2022, 12:16 AMThis is awesome, are you using pix2pix or is it a difference system? - u
{K EY1} (Kei)
06/30/2022, 12:16 AMI also wonder how the training goes Like do you need transcriptions of the target voice's lines, or no? - u
{K EY1} (Kei)
06/30/2022, 12:22 AMIsnt pix2pix for images - c
Cris140
06/30/2022, 12:24 AMYes, but you could use it with spectograms - p
Pathos
06/30/2022, 12:24 AMIt's a custom system from about, 4 modern VC papers - c
Cris140
06/30/2022, 12:25 AMThat's awesome, Good work