Why does not augmentation increase the number of training data? #1784
|
I have used vad configure file including augmenor with shift and white noise perturbation, spec_augment but when I try to apply it in train phase with 1000 training samples, the number of traning samples is fixed and it does not increase by augmentation. Why does not my training samples with these augmentations increase? |
Answered by
nithinraok
Feb 17, 2021
Replies: 1 comment
|
Perturbation augmentation in NeMo is applied if probability at random sample > prob of augmentation selected. So total num of samples (iterations) remains the same. I suggest you train for a higher number of epochs for effective usage of augmentation. |
0 replies
Answer selected by
okuchaiev
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Perturbation augmentation in NeMo is applied if probability at random sample > prob of augmentation selected. So total num of samples (iterations) remains the same. I suggest you train for a higher number of epochs for effective usage of augmentation.