espnet2.asr.postencoder.abs_postencoder.AbsPostEncoder

source

class espnet2.asr.postencoder.abs_postencoder.AbsPostEncoder(*args, **kwargs)

Bases: Module, ABC

Abstract base class for post-encoder modules in ASR (Automatic Speech Recognition) systems.

This class defines the interface for post-encoder components that process the output from encoders. Any specific implementation must inherit from this class and implement the abstract methods defined herein.

None

output_size() → int

Returns the size of the output tensor from the post-encoder.

forward(input

torch.Tensor, input_lengths: torch.Tensor) ->

Tuple()

Processes the input tensor and its corresponding lengths to produce the output tensor and updated lengths.

• Raises:
  • NotImplementedError – If the abstract methods are not implemented in
  • a subclass. –

####### Examples

To create a specific post-encoder, subclass this abstract class and implement the required methods:

``

`

python class MyPostEncoder(AbsPostEncoder):

def output_size(self) -> int: : return 256 # Example output size

def forward(self, input: torch.Tensor, input_lengths: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:

Custom processing logic here

return input, input_lengths # Example return

``

`

NOTE

This class should not be instantiated directly. It is meant to be subclassed to provide concrete implementations for various post-encoder functionalities.

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

abstract forward(input: Tensor, input_lengths: Tensor) → Tuple[Tensor, Tensor]

Returns the size of the output tensor produced by the encoder.

This method should be implemented by subclasses to provide the expected output size based on the specific encoding strategy.

• Returns: The size of the output tensor.
• Return type: int

abstract output_size() → int

Returns the size of the output produced by the encoder.

This method should be implemented by subclasses to provide the specific output size based on the encoding configuration. The output size is typically determined by the architecture of the encoder, which may vary based on the model design.

• Returns: The size of the output from the encoder.
• Return type: int
• Raises:
  • NotImplementedError – If the method is not implemented by a
  • subclass. –

####### Examples

>>> class MyEncoder(AbsPostEncoder):
...     def output_size(self) -> int:
...         return 256
...
>>> encoder = MyEncoder()
>>> encoder.output_size()
256