# machine-learning
- t
Tiege
02/14/2023, 4:00 AM - t
Tiege
02/14/2023, 4:30 AMI'm a little worried about creative communities (artists, voice actors, etc.) pushing for governments and companies to restrict AI because it might lead to unintended (and dystopian) consequences - u
(Dawn) Will Draw Fictional Women
02/14/2023, 4:35 AMi'm sure at least some of it is completely intended - t
Tiege
02/15/2023, 8:10 PM - p
PeaNutsAreGood
02/16/2023, 5:06 AMTortoise-TTS dvae model was just found, which should allow it to be finetuned now if anyone is interested https://github.com/152334H/DL-Art-School - i
I love the hive
02/16/2023, 10:56 AMI will use this for Tortoise TTS - i
I love the hive
02/16/2023, 11:34 AMresult: - j
Justin
02/16/2023, 11:48 AM - h
hecko
02/16/2023, 12:13 PMuberduck real?! - u
(Dawn) Will Draw Fictional Women
02/16/2023, 6:26 PMwhens this for mugen cns - c
ccateni
02/16/2023, 9:07 PMSo i've been toying with elevenlab and comparing it to Uberduck, found some interesting results. - c
ccateni
02/16/2023, 9:07 PMI think Uberduck's tacotron and hifi-gan have better detection of the VAs voice for that characters. - c
ccateni
02/16/2023, 9:08 PMElevenlabs like to get past the character's additions and gets to the source of the voice resulting in the characters sound wrong. - c
ccateni
02/16/2023, 9:10 PMheres an example. Heres a rough test from Uberduck using my model (ignore the roughness, I'm most likely going to do more training with the model to fix the issue) - c
ccateni
02/16/2023, 9:11 PMand here is the elevenlabs attempt at Suketoudara using the same clips (only 25 of them) from the model I created. - c
ccateni
02/16/2023, 9:11 PMThis isn't the only time this happened either - c
ccateni
02/16/2023, 9:13 PMHeres the werehog model from Uberduck saying "Stop calling me Sonic, ok? Its the werehog. Don't ask why I call myself that." (again, this one needs retraining, so ignore the model having blunders) - c
ccateni
02/16/2023, 9:14 PMHeres the Elevenlabs attempt. - c
ccateni
02/16/2023, 9:14 PMThe elevenlabs sounds more like Jason's normal voice rather than the voice he used for the werehog, same with Suketoudara. - c
ccateni
02/16/2023, 9:14 PMIts a common theme. - u
(Dawn) Will Draw Fictional Women
02/16/2023, 9:15 PMthat's zeroshot! - c
ccateni
02/16/2023, 9:18 PMZeroshot? - c
ccateni
02/16/2023, 9:18 PMas in Zeroshot learning? - h
hecko
02/16/2023, 9:18 PMyeah - c
ccateni
02/16/2023, 9:19 PMYou think its the samples or the AI elevenlabs uses? - h
hecko
02/16/2023, 9:19 PMit's probably part because their data is audiobooks and part because zeroshot can only go so far in general - c
ccateni
02/16/2023, 9:19 PMinteresting.