subsampling

Module Contents

Classes

BaseSubsampling
Conv2dSubsampling4	Convolutional 2D subsampling (to 1/4 length).

Attributes

pos

class subsampling.BaseSubsampling

Bases: tensorflow.keras.layers.Layer

position_encoding(offset: int, size: int) → tensorflow.Tensor

compute_logit_length(input_length)

used for get logit length

class subsampling.Conv2dSubsampling4(num_filters: int, odim: int, feats_dim: int, dropout_rate: float, pos_enc_class: tensorflow.keras.layers.Layer)

Bases: BaseSubsampling

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

Parameters

• idim (int) – Input dimension.

• odim (int) – Output dimension.

• dropout_rate (float) – Dropout rate.

call(input: tensorflow.Tensor, x_mask: tensorflow.Tensor, offset: int = 0) → Tuple[Any, Any, Any]

Subsample x.

Parameters

• x (tf.Tensor) – Input tensor (#batch, time, idim, 1).

• x_mask (tf.Tensor) – Input mask (#batch, 1, 1, time).

Returns

Subsampled tensor (#batch, time’, odim, 1),

where time’ = time // 4.

torch.Tensor: Subsampled mask (#batch, 1, time’),

where time’ = time // 4.

torch.Tensor: positional encoding

Return type

torch.Tensor

compute_logit_length(input_length)

used for get logit length

subsampling.pos