espnet.nets.pytorch_backend.conformer.contextual_block_encoder_layer.ContextualBlockEncoderLayer

source

class espnet.nets.pytorch_backend.conformer.contextual_block_encoder_layer.ContextualBlockEncoderLayer(size, self_attn, feed_forward, feed_forward_macaron, conv_module, dropout_rate, total_layer_num, normalize_before=True, concat_after=False)

Bases: Module

Contexutal Block Encoder layer module.

• Parameters:
  • size (int) – Input dimension.
  • self_attn (torch.nn.Module) – Self-attention module instance. MultiHeadedAttention or RelPositionMultiHeadedAttention instance can be used as the argument.
  • feed_forward (torch.nn.Module) – Feed-forward module instance. PositionwiseFeedForward, MultiLayeredConv1d, or Conv1dLinear instance can be used as the argument.
  • feed_forward_macaron (torch.nn.Module) – Additional feed-forward module instance. PositionwiseFeedForward, MultiLayeredConv1d, or Conv1dLinear instance can be used as the argument.
  • conv_module (torch.nn.Module) – Convolution module instance. ConvlutionModule instance can be used as the argument.
  • dropout_rate (float) – Dropout rate.
  • total_layer_num (int) – Total number of layers
  • normalize_before (bool) – Whether to use layer_norm before the first block.
  • concat_after (bool) – Whether to concat attention layer’s input and output. if True, additional linear will be applied. i.e. x -> x + linear(concat(x, att(x))) if False, no additional linear will be applied. i.e. x -> x + att(x)

Construct an EncoderLayer object.

forward(x, mask, infer_mode=False, past_ctx=None, next_ctx=None, is_short_segment=False, layer_idx=0, cache=None)

Calculate forward propagation.

forward_infer(x, mask, past_ctx=None, next_ctx=None, is_short_segment=False, layer_idx=0, cache=None)

Compute encoded features.

• Parameters:
  • x_input (torch.Tensor) – Input tensor (#batch, time, size).
  • mask (torch.Tensor) – Mask tensor for the input (#batch, 1, time).
  • past_ctx (torch.Tensor) – Previous contexutal vector
  • next_ctx (torch.Tensor) – Next contexutal vector
  • cache (torch.Tensor) – Cache tensor of the input (#batch, time - 1, size).
• Returns: Output tensor (#batch, time, size). torch.Tensor: Mask tensor (#batch, 1, time). cur_ctx (torch.Tensor): Current contexutal vector next_ctx (torch.Tensor): Next contexutal vector layer_idx (int): layer index number
• Return type: torch.Tensor

forward_train(x, mask, past_ctx=None, next_ctx=None, layer_idx=0, cache=None)

Compute encoded features.

• Parameters:
  • x_input (torch.Tensor) – Input tensor (#batch, time, size).
  • mask (torch.Tensor) – Mask tensor for the input (#batch, time).
  • past_ctx (torch.Tensor) – Previous contexutal vector
  • next_ctx (torch.Tensor) – Next contexutal vector
  • cache (torch.Tensor) – Cache tensor of the input (#batch, time - 1, size).
• Returns: Output tensor (#batch, time, size). torch.Tensor: Mask tensor (#batch, time). cur_ctx (torch.Tensor): Current contexutal vector next_ctx (torch.Tensor): Next contexutal vector layer_idx (int): layer index number
• Return type: torch.Tensor