espnet2.asr.state_spaces.components.ReversibleInstanceNorm1dInput

source

class espnet2.asr.state_spaces.components.ReversibleInstanceNorm1dInput(d, transposed=False)

Bases: Module

Reversible Instance Normalization for 1D inputs.

This module implements reversible instance normalization for 1D inputs. It computes the mean and standard deviation of the input tensor and normalizes the input accordingly. The normalization parameters can be reversed to retrieve the original input.

transposed

Indicates whether the input is in transposed form (BDL) or not (BLD).

• Type: bool

norm

Instance normalization layer.

• Type: nn.InstanceNorm1d

• Parameters:

  • d (int) – The number of features in the input tensor.
  • transposed (bool , optional) – If True, expects input shape (B, D, L). Default is False (input shape is (B, L, D)).

• Returns: The normalized input tensor.

• Return type: Tensor

####### Examples

>>> norm_layer = ReversibleInstanceNorm1dInput(d=10, transposed=False)
>>> input_tensor = torch.randn(32, 5, 10)  # (Batch, Length, Features)
>>> normalized_tensor = norm_layer(input_tensor)
>>> print(normalized_tensor.shape)
torch.Size([32, 5, 10])

NOTE

This module is designed to work with transposed and non-transposed inputs, allowing flexibility in handling different data formats.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x)

ReversibleInstanceNorm1dInput class.

This class applies a reversible instance normalization operation over the input tensor. It computes the mean and standard deviation along the specified dimensions and normalizes the input accordingly. The normalization is reversible, allowing the original input to be reconstructed later using the stored statistics.

transposed

A flag indicating whether the input tensor is in a transposed format (BDL) or not (BLD).

• Type: bool

norm

The instance normalization layer.

• Type: nn.InstanceNorm1d

• Parameters:

  • d (int) – The number of features in the input tensor.
  • transposed (bool) – Indicates if the input is in a transposed format. Default is False.

• Returns: The normalized tensor, with the same shape as the input.

• Return type: Tensor

####### Examples

>>> layer = ReversibleInstanceNorm1dInput(d=64, transposed=False)
>>> input_tensor = torch.randn(32, 64, 10)  # (batch_size, features, lengths)
>>> output_tensor = layer(input_tensor)
>>> print(output_tensor.shape)  # Output: torch.Size([32, 64, 10])

NOTE

This normalization is particularly useful in scenarios where maintaining the input distribution is critical, such as in reversible networks.