# machine-learning
- c
Cris140
09/26/2022, 4:46 AMPerfect, thank you so much 🙏 - d
Dalton
09/27/2022, 2:01 AMdoes anyone know if theres anysites using this ? https://github.com/neonbjb/tortoise-tts - d
Dalton
09/27/2022, 2:01 AMthey samples sound REALLY REAL - u
(Dawn) Will Draw Fictional Women
09/27/2022, 2:01 AMyou cant implement this into a site - u
(Dawn) Will Draw Fictional Women
09/27/2022, 2:01 AMit requires gpu to work with good speed which is super costly - d
Dalton
09/27/2022, 2:02 AMis this not a model type like tacotron or talknet? - u
(Dawn) Will Draw Fictional Women
09/27/2022, 2:03 AMno - d
Dalton
09/27/2022, 2:03 AMoh I thought it was - d
Dalton
09/27/2022, 2:03 AMim stupid srry - h
hecko
09/27/2022, 8:08 AMit's closer to 15.ai in that it has a bunch of voices crammed into one giant model so many in fact that it can mix them together to sorta emulate any voice you give it but unfortunately it's also extremely slow, takes like a minute to generate one audio file on colab on default settings - h
hecko
09/27/2022, 8:08 AMplus since it was trained on podcasts and such it's not very good at the wacky cartoon voices we have at uberduck - a
a cat came into my house
09/27/2022, 11:50 PMhi, new pipeline trainer here. am i supposed to use the "wavs/" format for transcriptions txts? (like in legacy training) if this is not in the right channel then let me know - r
real sky 2022 (how)
09/27/2022, 11:51 PMthe preferred channel is #994486394049282058 but its okay yes, the formatting is the same - f
fatherallah
09/28/2022, 2:31 AMAnyone here successfully make a great singing model on talknet? With the dataset composed of a singer’s acapellas? My first one came out awful and I want to know if I should keep trying or find an alternative. - p
postmates!!
09/28/2022, 3:37 AMwell that was easier then i expected - a
a cat came into my house
09/28/2022, 5:07 AMtraining a model of the "Uh oh, you found the toothpaste!" guy. i think he sounds good - a
a cat came into my house
09/28/2022, 5:12 AMeven though i might have to retrain him, which takes hours all because i forgot his "You found the toothpaste" line - a
a cat came into my house
09/28/2022, 5:34 AMfuck it. retraining just to add the toothpaste lines - a
a cat came into my house
09/28/2022, 7:00 AMalright it's time for the FFmpeg arpabet test. - a
a cat came into my house
09/28/2022, 7:00 AMsmall passed the test... kind of! - p
postmates!!
09/28/2022, 2:20 PMgot some meh results from my edd model - p
postmates!!
09/28/2022, 2:21 PMsounds like it needs more time in the oven - h
HolyArapaima
09/28/2022, 4:49 PMThis is the best singing model I got to output out of talknet, - c
Couch
09/28/2022, 4:49 PMif i recall someone said consistency is a huge factor when it comes to talknet - h
HolyArapaima
09/28/2022, 4:49 PMYes - h
HolyArapaima
09/28/2022, 4:49 PMIt's unpredictable - h
HolyArapaima
09/28/2022, 4:50 PMYou can skew your learning rate and stuff to help you out but it's very trial and error - h
HolyArapaima
09/28/2022, 4:51 PMAfter I got this output out of talknet I struggled to get the model to replicate it lol - h
HolyArapaima
09/28/2022, 4:51 PMI found that it obviously worked the best using words and singing phrases already recorded obviously