# machine-learning
- u
(Dawn) Will Draw Fictional Women
09/25/2021, 1:04 AMonly input the id between the slashes - u
user
09/25/2021, 1:04 AMalright - u
user
09/25/2021, 1:04 AMthanks - u
user
09/25/2021, 1:23 AMthanks, it works! - c
cengi
09/25/2021, 10:01 AMif i close the tab while training, does it close the training or does it continue to train on my account colab ? - h
hecko
09/25/2021, 10:22 AMit stops the training after a few minutes yeah - c
cengi
09/25/2021, 12:28 PMwhy is it not recommended to change the total epochs to train to ? - h
hecko
09/25/2021, 12:41 PMi'd imagine it's because many people would try training for way too long and mess up their models (called "overfitting") - c
cengi
09/25/2021, 1:30 PMok thanks - c
cengi
09/25/2021, 1:31 PMHow do I add new voice samples when it's already done training ? - h
hecko
09/25/2021, 1:32 PMjust start training with the same settings and it'll resume - h
hecko
09/25/2021, 1:32 PMthe epoch count will be slightly off because it's calculated from the saved iteration count - c
cengi
09/25/2021, 1:33 PMDo I import all the previous wav samples used too ? how about lists txt file ? - h
hecko
09/25/2021, 1:33 PMyes and yes - h
hecko
09/25/2021, 1:34 PMalthough personally i'd just restart training if possible - h
hecko
09/25/2021, 1:34 PMbecause training on a low amount of samples does degrade the model's speech capabilities - c
cengi
09/25/2021, 1:34 PMi see.... - h
hecko
09/25/2021, 1:49 PMwhile you're at it try this notebook https://colab.research.google.com/drive/1PwYVOYWRXX7itYB2lKYOhstyBySz8DBq?usp=sharing it trains the model such that it's compatible with the pronunciation_dictionary option, which means better pronunciation - c
cengi
09/25/2021, 1:51 PMdoes it taking longer to train than the normal one ? i might try it - h
hecko
09/25/2021, 1:54 PMit takes a while longer to boot up but training is the same speed - u
user
09/25/2021, 5:54 PMIs the tutorial video by Cherry Studios still considered up to date? Or has the process of training changed considerably by now? - c
ColdFir3
09/25/2021, 8:13 PMHow comes this notebook isn't listed in the #841437191073955920 section - h
hecko
09/25/2021, 8:13 PM¯\_(ツ)_/¯ - c
ColdFir3
09/25/2021, 8:14 PMMaybe one day I'll implement it into my kaggle notebook - h
hecko
09/25/2021, 8:44 PMwhile you're at it you can add my dataset loader thing https://colab.research.google.com/drive/1xnFT5gXY5CIdsyG0YZbCulXdJksJR4ei - h
hecko
09/25/2021, 8:44 PMthough i've heard kaggle has persistent storage, so it might either be less useful or at least would benefit from being adjusted to load into there - e
Ern851
09/25/2021, 9:19 PMI have a question... I'm doing a voice ai that has around 70 to 80 wav files, each around 5 to 10 seconds long, and I'm wondering what is the right batch size and the right number of epochs I could use so that the audio can be trained well. - u
(Dawn) Will Draw Fictional Women
09/25/2021, 9:20 PMim not claiming this is correct but generally i try to stretch my batch size as much as i can - u
(Dawn) Will Draw Fictional Women
09/25/2021, 9:21 PMas for number of epochs i try to make that pretty large as well cause i usually manually stop my training when i hit a good validation loss