add sequence masking for convolution ops

NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP

https://nvidia.github.io/OpenSeq2Seq

Apache License 2.0

1.54k stars 369 forks source link

Closed blisc closed 5 years ago

blisc commented 5 years ago

Add option for using a sequence mask prior to convolution operations. To enable, set 'use_conv_mask': True inside the encoder.

Signed-off-by: Jason jasoli@nvidia.com