train_CNN about subtitle-synchronization HOT 4 CLOSED

Hi! It should be trained with the train_dense function. However, model_dense is the recommended one for this problem, since it includes also 1-D Convolutional layers.

from subtitle-synchronization.

Thanks, but I still don't understand why len_mfcc==1 and step_mfcc==1 like picture below,I think it should be len_mfcc!=1 and step_mfcc!=1
Can you explain me more
https://drive.google.com/file/d/1GXXHxYAMZ7vtg11bLQUt7K4dnvjZij3d/view?usp=sharing

from subtitle-synchronization.

len_mfcc == 1 create training samples of the shape (len_mfcc, #fcc_features)
len_mfcc > 1 create samples with more than one timesteps
step_mfcc is used to create training samples with overlapping between them

from subtitle-synchronization.

I understand the following:
len_mfcc == 1 for train CNN
len_mfcc>1 for train RNN
function train_lstm means train according to RNN model and function train_dense means train according to CNN model, right?
If I have a mistake or misunderstanding please help

from subtitle-synchronization.