# machine-learning
- t
The Wiggles and High School D×D
11/15/2021, 5:08 PMYeah, I was mentioning I was using Python to this guy He told me for complex programs a C language would be more efficient for the computer to process - d
Desuka
11/15/2021, 5:08 PMHow many voice clips did it have approximately? - c
Couch
11/15/2021, 5:10 PMi guess that's true since c compiles directly to machine code whereas python is just interpreted - c
Couch
11/15/2021, 5:11 PMthe computer takes time running the vm and the code - t
The Wiggles and High School D×D
11/15/2021, 5:11 PMIt was something I considered looking into if only because I'm still stuck on my bilingual voice cloning thing And I probably would need to dig into the code and maybe overhaul the thing - t
The Wiggles and High School D×D
11/15/2021, 5:11 PMBut I don't know - t
The Wiggles and High School D×D
11/15/2021, 5:11 PMI keep thinking I won't utilise the language enough - h
hecko
11/15/2021, 5:58 PMafaik pytorch and such do use c in the backend - h
hecko
11/15/2021, 5:59 PMspecifically due to efficiency - m
mega b
11/15/2021, 7:44 PMquantanizing tacotron2 models (8bit) could be fun to try - m
mega b
11/15/2021, 7:46 PMe.g inferencing on your phone cpu - u
user
11/15/2021, 9:24 PMwhat even happens if you go above the recommended epoch amount - u
(Dawn) Will Draw Fictional Women
11/15/2021, 9:26 PMincreased chance of overfitting - c
Couch
11/15/2021, 9:30 PMlower the training rate - t
The Wiggles and High School D×D
11/16/2021, 12:19 AMTaking another crack at the machine. In the step to preprocess the synthesiser data I get this error. I can't quite figure out the issue but I want to assume it has to do with me not being able to compile data from the encoder step somehow? I can't tell. - u
user
11/16/2021, 12:22 PM30 epoch trained so far - u
user
11/16/2021, 2:37 PMHi, I tried to train my own voice but all it's outputting is silence. The voice has been through a total of about 500(?)-ish epochs and uses 112 22kHz Mono
files..wav - u
user
11/16/2021, 2:48 PMOkay nvm, I think I figured it out; I was using the Colab from Adam's tutorial - u
user
11/16/2021, 2:48 PMFor synthesis - u
user
11/16/2021, 2:58 PMIt sounds bad. I might need to re-train it later. - u
user
11/16/2021, 2:58 PMNot right now though as class is gonna start soon - u
user
11/16/2021, 4:11 PMOkay so right before class started I was able to resume training, this time via the correct Colab - m
matty1406
11/16/2021, 5:44 PMis there any good learning rate numbers or certain epochs i should use for the talknet notebook to try make the model sound better? - m
matty1406
11/16/2021, 5:45 PMjust hoping aint working anymore... - u
(Dawn) Will Draw Fictional Women
11/16/2021, 6:03 PMeveryone just keeps telling me “leave everything alone” but im not buyin it - u
(Dawn) Will Draw Fictional Women
11/16/2021, 6:04 PMthere’s gotta be shit i can change to make things better - u
(Dawn) Will Draw Fictional Women
11/16/2021, 6:04 PMtry increasing the epochs on spectrogram to a ridiculous amount and stop it manually when you get consistent not in top 3s - m
mega b
11/16/2021, 6:12 PMhyperperameter search - c
CerealYe
11/16/2021, 6:14 PM*kid