i never fuck with it

I set learning rate to 0.0004 for training with 4 wavs that are approx 10 seconds long and it works fine.

help

@hecko Can you identify the cause of this error?

uhhhhhhhhh

depends on the output of step 2

All you gotta do is

This might be a really stupid question to all of you, but I'd like to inquire what training and validation filelists are?

How many audio files should I aim for, when training a voice model of my own voice?

as much as humanly possible

is 113 going to be sufficient

measure by length not file count

hm

that's around 3 minutes of dialogue

ImportError Traceback (most recent call last) in 10 11 import torch ---> 12 from distributed import apply_gradient_allreduce 13 import torch.distributed as dist 14 from torch.utils.data.distributed import DistributedSampler ImportError: cannot import name 'apply_gradient_allreduce' from 'distributed' (/usr/local/lib/python3.8/dist-packages/distributed/__init__.py) uhmmm How

in machine learning people are Supposed to separate out some of the data into a validation set that's only used for scoring and not training, so that when the validation loss is low you know it's because the network is actually good and not because it memorized the training data really well (which is called overfitting) we don't do that though, because: a) we often have so little data that not training on some of it would be a waste b) people actually prefer the sound of overfit models c) the original notebook creator didn't feel like making a proper validation splitter and everyone else followed d) all of the above

okay thanks but

what about the error i got yesterday

okay it's probably fixed but i have another error

AttributeError Traceback (most recent call last) in 375 376 # ---- DEFAULT PARAMETERS DEFINED HERE ---- --> 377 hparams = create_hparams() 378 model_filename = 'current_model' 379 hparams.training_files = "filelists/clipper_train_filelist.txt" /content/tacotron2/hparams.py in create_hparams(hparams_string, verbose) 6 super(HParamsAlternative, self).__init__(*args, **kwargs) 7 self.__dict__ = self ----> 8 9 def create_hparams(hparams_string=None, verbose=False): 10 """Create model hyperparameters. Parse nondefault from given string."""

you didn't paste the actual error

but that seems like a very weird step to error on

before anything, make sure you're using the most updated notebook on #841437191073955920

How many hours of training does it usually take?

On a T4

How much do you usually use?

just that

as much as humanly possible

…throttled by me being lazy occasionally

Remember, quality > quantity