espnet2.enh.encoder.null_encoder.NullEncoder

source

class espnet2.enh.encoder.null_encoder.NullEncoder

Bases: AbsEncoder

NullEncoder is a simple implementation of an encoder that performs no

operations on the input. It is a subclass of AbsEncoder and serves as a placeholder encoder.

output_dim

The output dimension of the encoder, which is fixed at 1.

• Type: int

• Parameters:

  • input (torch.Tensor) – Mixed speech tensor of shape [Batch, sample].
  • ilens (torch.Tensor) – Tensor representing the lengths of the input sequences, shape [Batch].
  • fs (int , optional) – Sampling rate in Hz. This parameter is not used in this encoder.

• Returns: A tuple containing: : - torch.Tensor: The input tensor unchanged.

  • torch.Tensor: The input lengths unchanged.

• Return type: tuple

######### Examples

>>> encoder = NullEncoder()
>>> mixed_speech = torch.randn(5, 16000)  # Example mixed speech
>>> input_lengths = torch.tensor([16000] * 5)  # Example input lengths
>>> output, lengths = encoder.forward(mixed_speech, input_lengths)
>>> assert torch.equal(output, mixed_speech)
>>> assert torch.equal(lengths, input_lengths)

NOTE

This encoder is primarily used for testing or when no processing is required.

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

#

forward(input

: Tensor, ilens: Tensor, fs: int | None = None)

Forward pass of the NullEncoder.

This method takes the mixed speech input and its corresponding lengths, and returns them without any modifications. It is primarily used in scenarios where an encoder is required, but no actual processing is needed. This is useful for testing or as a placeholder in models.

• Parameters:
  • input (torch.Tensor) – Mixed speech tensor of shape [Batch, sample].
  • ilens (torch.Tensor) – Tensor containing the lengths of each input sequence, shape [Batch].
  • fs (int , optional) – Sampling rate in Hz (Not used). Default is None.
• Returns: Returns the input tensor and the : input lengths tensor as a tuple.
• Return type: Tuple[torch.Tensor, torch.Tensor]

######### Examples

>>> encoder = NullEncoder()
>>> mixed_speech = torch.randn(5, 16000)  # 5 samples, 16000 samples each
>>> lengths = torch.tensor([16000, 16000, 16000, 16000, 16000])
>>> output, output_lengths = encoder.forward(mixed_speech, lengths)
>>> output.shape
torch.Size([5, 16000])
>>> output_lengths.shape
torch.Size([5])

NOTE

This encoder does not perform any transformations on the input data. It simply returns the inputs as they are.

property output_dim : int

Null encoder.

This encoder serves as a placeholder and does not perform any actual encoding. It simply returns the input as is.

output_dim

The dimensionality of the output, which is always 1.

• Type: int

forward(input

torch.Tensor, ilens: torch.Tensor, fs: int = None) -> Tuple[torch.Tensor, torch.Tensor]: Passes the input tensor and its lengths through without modification.

######### Examples

encoder = NullEncoder() print(encoder.output_dim) # Output: 1

input_tensor = torch.tensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]) ilens_tensor = torch.tensor([3, 3]) output, lengths = encoder.forward(input_tensor, ilens_tensor) print(output) # Output: tensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]) print(lengths) # Output: tensor([3, 3])