espnet.nets.chainer_backend.transformer.subsampling.Conv2dSubsampling

source

class espnet.nets.chainer_backend.transformer.subsampling.Conv2dSubsampling(channels, idim, dims, dropout=0.1, initialW=None, initial_bias=None)

Bases: Chain

Convolutional 2D subsampling (to 1/4 length).

• Parameters:
  • idim (int) – input dim
  • odim (int) – output dim
  • dropout_rate (flaot) – dropout rate

Initialize Conv2dSubsampling.

forward(xs, ilens)

Subsample x.

• Parameters:x (chainer.Variable) – input tensor
• Returns: subsampled x and mask