# machine-learning
- u
(Dawn) Will Draw Fictional Women
08/26/2022, 8:49 PM> [W pthreadpool-cpp.cc:90] Warning: Leaking Caffe2 thread-pool after fork. (function pthreadpool) - u
(Dawn) Will Draw Fictional Women
08/26/2022, 8:49 PM>appears when training custom hifigan - u
(Dawn) Will Draw Fictional Women
08/26/2022, 8:50 PMwhat mean - u
(Dawn) Will Draw Fictional Women
08/26/2022, 8:50 PMshould me be scare? - u
(Dawn) Will Draw Fictional Women
08/26/2022, 8:50 PMis bad happen? - m
mega b
08/26/2022, 11:10 PMIt means the pipe connecting to Starbucks has been damaged - m
mega b
08/26/2022, 11:10 PMIt made a pool - t
THE UBERSHIT
08/26/2022, 11:47 PMi'm thinking i should try to train models other than legacy tacotron. i wonder how that would turn out - j
Justin
08/27/2022, 2:22 AM - z
zwf
08/27/2022, 5:07 PMhonestly so cool - u
(Dawn) Will Draw Fictional Women
08/27/2022, 5:12 PMis that a stable diffusion plugin for photoshop??? - j
Justin
08/27/2022, 5:15 PMYup - j
JamesVincentWalsh
08/27/2022, 11:40 PMFinally decided to update the first model I trained so it has more audio to learn from. Not sure if it's going well or going terribly, but it sure is going. - m
mega b
08/28/2022, 12:39 AM - m
mega b
08/28/2022, 12:39 AMcc @samkoelle this might be cool - s
samkoelle
08/28/2022, 12:40 AMvery cool - s
samkoelle
08/28/2022, 12:41 AMyeah anything based on taco is nice since ultimately we are just gonna hack the shit out of the taco architecture until its achieved its platonic ideal - s
samkoelle
08/28/2022, 12:41 AMor at least, thats the plan - m
mega b
08/28/2022, 12:41 AMdope - j
JamesVincentWalsh
08/28/2022, 12:42 AMMakes me wonder how many decades it'll be before we finally have natural-sounding tts that you can customize to sound however you want... - s
samkoelle
08/28/2022, 12:42 AMhonestly wouldnt be that surprised if its in the harmonai release next month - s
samkoelle
08/28/2022, 12:43 AM5 years max - s
samkoelle
08/28/2022, 12:43 AMmaybe im overly optimistic hah - m
mega b
08/28/2022, 12:44 AMi think its a good duration but tbh development in tts in general is not as active as other fields nowadays (e.g visuals & images) - j
JamesVincentWalsh
08/28/2022, 12:46 AMokay yeah it's struggling Could it be because of batch size? I'm training it on 155 wavs and set the batch size to 4, but a quick google search has now told me that 5 divides evenly 🤔 - j
JamesVincentWalsh
08/28/2022, 12:46 AMYeah it feels like people are so distracted by visuals now that nobody really cares much about tts anymore - s
samkoelle
08/28/2022, 12:49 AMyou know what they say - ends with 5, divides by 5 - j
JamesVincentWalsh
08/28/2022, 12:52 AMI've never heard that before, but that is incredibly useful to know - j
JamesVincentWalsh
08/28/2022, 1:38 AMThink I should stop it at a graph like this, or let it keep going? - j
JamesVincentWalsh
08/28/2022, 2:24 AMDecided to call it. The new version somehow sounds worse than the old version e.o