Output size is weird

[kuraga]

On the README example,

inputs.size()  # => torch.Size([3, 1234, 80])
output.size()  # => torch.Size([3, 556, 10])

How do I save the time-dimension?

[kuraga]

Seems like the first block in the encoder and the block of the decoder should both be 1-block instead. This change makes calculate padding instead of getting it to 1.

NVIDIA's DeepLearningExamples JASPER helped me to get this: 1, 2.