# machine-learning
- i
IBob012
06/04/2021, 5:42 AMi mean you can train the vocoder to only one character but training on multiple is probably better since its one vocoder for all of them instead of multiple - i
IBob012
06/04/2021, 5:44 AMgoing to sleep, hit me up later to talk more - s
SuperLuig
06/04/2021, 5:50 AMye - w
watiba
06/04/2021, 7:34 AM@IBob012 need Hifi-Gan colab link. - u
user
06/04/2021, 7:46 AMHes sleep - m
Marconan
06/04/2021, 9:59 AMWouldn't it make more sense to use something like the Dialog De-reverb neural net in Izotope RX8 than training it on reverb'd data? - u
user
06/04/2021, 11:43 AMyou can spot the alamy stock logo here - b
Balloonya from Pokémon
06/04/2021, 1:49 PMI've 116 audio files of Ozora Tsubasa (Captain Tsubasa 2002). I'll put track name or convert all files? - a
Awesome Face (GTX 1660 Ti)
06/04/2021, 1:53 PMjust check convertall and you are good to go - b
Balloonya from Pokémon
06/04/2021, 1:54 PMokay - b
Balloonya from Pokémon
06/04/2021, 2:34 PMIn Vocal Remover 5 notebook, downloading files with convertAll won't work yet. - b
Balloonya from Pokémon
06/04/2021, 2:41 PMIt works! - s
SuperLuig
06/04/2021, 3:15 PMsnugglemagne: has the same amount of data as tanya and eugene a consistent tone of voice (unlike eugene which negatively affects his model) so why did snugglemagne turn out so bad - s
SuperLuig
06/04/2021, 3:15 PMthe eugene voice sounds like shit but it at least works - s
SuperLuig
06/04/2021, 3:18 PMis it the way snugglemagne talks/his voice in general? - s
SuperLuig
06/04/2021, 3:19 PMlike, would y'all consider this hard to train? - s
SuperLuig
06/04/2021, 3:19 PMi cant provide any more examples cause i'm not home rn - s
SuperLuig
06/04/2021, 3:22 PMthere's one more episode where he has a major role that I could get clips from - s
SuperLuig
06/04/2021, 3:32 PMliterally this - m
mega b
06/04/2021, 3:35 PMthe models you want the most, comes out as shit - m
mega b
06/04/2021, 3:36 PMthe models you dont care about, comes out as god laid his blessings over your colab - s
SuperLuig
06/04/2021, 3:44 PMi mean I theoretically have a solution to fix Adorabat's voice now that I know how to do it - s
SuperLuig
06/04/2021, 3:45 PMibob made a hifigan model of characters with reverb and it turned out great, using 500 lines in the dataset - s
SuperLuig
06/04/2021, 3:45 PMso if I make a hifigan model of high-pitched characters, that SHOULD work in theory, right? - s
SuperLuig
06/04/2021, 3:46 PMthere's plenty of datasets I could use here - s
SuperLuig
06/04/2021, 3:46 PMbubbles and someone else I forgot the name of would work especially well - b
Balloonya from Pokémon
06/04/2021, 3:46 PMIt's for use separated files to convert it into mono 22050? - s
SuperLuig
06/04/2021, 3:49 PMI wonder if I could cheat this by manually raising the pitch of certain characters in Audacity 🤔 - s
Scramble
06/04/2021, 3:50 PMThank god amy lee (the person who voiced coach) didn't have a very british accent