# machine-learning
- u
(Dawn) Will Draw Fictional Women
11/14/2021, 8:28 PMno - h
hecko
11/14/2021, 8:28 PMhttps://c.tenor.com/vw7ogSgBWuYAAAAM/no-yes.gif▾
- d
Desuka
11/14/2021, 8:28 PMOkay fine maybe it's a bit straining on the eyeballs. - h
hecko
11/14/2021, 8:28 PMmy chad is larger therefore i win - u
(Dawn) Will Draw Fictional Women
11/14/2021, 8:29 PMsee but youre not the one saying youre not gonna stop poking fun at what other people like (cringe) - u
(Dawn) Will Draw Fictional Women
11/14/2021, 8:29 PMits very simple - s
SpamZow (WoomyPearl)
11/14/2021, 11:36 PMNow updating Annoying Orange as it reached the 2K voice clip mark despite not sounding that good. - s
SpamZow (WoomyPearl)
11/14/2021, 11:37 PMAdded voice clips from few more videos from 2009 and 2010. - s
SpamZow (WoomyPearl)
11/14/2021, 11:38 PMPlus transcribed his laugh as someone asked me if he can laugh when I first released his model. - u
user
11/15/2021, 12:11 AM2 thousand voice clips?? - u
user
11/15/2021, 12:11 AMhow long does that take to train? 🤯 - b
Balloonya from Pokémon
11/15/2021, 11:07 AMI can't generate ground truth-aligned spectrograms for HiFi-GAN model. How to fix? - h
haru0l
11/15/2021, 11:22 AMi believe its due to nemo being updated - h
haru0l
11/15/2021, 11:22 AMwell the models anyway - b
Balloonya from Pokémon
11/15/2021, 11:24 AMIf the HiFiGAN Fine-Tune colab notebook will be updated, it will be fixed. - h
hecko
11/15/2021, 11:27 AMi believe he meant the model was used 2 thousand times - h
hecko
11/15/2021, 11:28 AMand idk about tacotron but my 1.5k-clip talknet model stopped improving after several hours - c
Couch
11/15/2021, 12:54 PMhifigan uses tacotron2 to generate ground truth spectrograms so you have to do the same edits on the tacotron repo to this notebook - c
Couch
11/15/2021, 12:54 PMthat means changing symbols.py - c
Couch
11/15/2021, 12:54 PMand ignoring embedding weight layer (i think) - t
Traincraft101
11/15/2021, 3:51 PMIs there any benefit to training above 2000 steps on "Train HiFi-GAN" talknet? - u
(Dawn) Will Draw Fictional Women
11/15/2021, 3:52 PMits been known to improve some models - u
(Dawn) Will Draw Fictional Women
11/15/2021, 3:52 PMi recall one of jacoblenstar’s talknets not sounding good at 2000 but drastically improving at 6000 - j
JacobLenstar
11/15/2021, 3:53 PMyeah when I don't get the CUDA error - u
(Dawn) Will Draw Fictional Women
11/15/2021, 3:53 PMcuda issue - t
The Wiggles and High School D×D
11/15/2021, 4:55 PMSo there's this guy I'm talking to And he's telling me about how good C languages (C, C#, C++) are good for AI I feel as if this statement has some kind of application to our research into voice cloning but I don't know what kind of potential is untapped here or how to address it, it's not my place to say - c
Couch
11/15/2021, 5:03 PMI've seen python being used very often for ML (see pytorch) - c
Couch
11/15/2021, 5:04 PMthe only cpp thing i know that has to do with voice cloning is tensorvox (which uses a c port of tensorflow i believe)