espnet2.diar.decoder.abs_decoder.AbsDecoder

source

class espnet2.diar.decoder.abs_decoder.AbsDecoder(*args, **kwargs)

Bases: Module, ABC

Abstract base class for decoders in the ESPnet2 diarization module.

This class serves as a blueprint for all decoder implementations, enforcing the implementation of the forward method and the num_spk property in derived classes. It inherits from torch.nn.Module and implements the necessary interfaces for building neural network models.

num_spk

The number of speakers, must be implemented in derived classes.

• Type: int

forward(input

torch.Tensor, ilens: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]: Defines the computation performed at every call.

• Parameters:
  • input (torch.Tensor) – The input tensor representing the data to be processed by the decoder.
  • ilens (torch.Tensor) – A tensor representing the lengths of the input sequences.
• Returns: A tuple containing the output tensor and any additional information (e.g., lengths or masks).
• Return type: Tuple[torch.Tensor, torch.Tensor]
• Raises:NotImplementedError – If the forward method or num_spk property is not implemented in the derived class.

######### Examples

class MyDecoder(AbsDecoder): : @property def num_spk(self): <br/>

return 2 <br/> def forward(self, input, ilens): : # Implementation of the forward method pass

decoder = MyDecoder() print(decoder.num_spk) # Output: 2

NOTE

This class should not be instantiated directly. Instead, create subclasses that implement the abstract methods.

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

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

Defines the forward pass for the AbsDecoder class, which is an abstract base class for decoders in a neural network model. This method must be implemented by any subclass of AbsDecoder to define the specific decoding mechanism.

• Parameters:
  • input (torch.Tensor) – The input tensor representing the encoded features that need to be decoded.
  • ilens (torch.Tensor) – A tensor representing the lengths of the input sequences. This is used to handle variable-length inputs.
• Returns: A tuple containing two tensors. The : first tensor is the decoded output, and the second tensor is a tensor representing any additional information required by the decoder (e.g., attention weights or hidden states).
• Return type: Tuple[torch.Tensor, torch.Tensor]
• Raises:NotImplementedError – If the method is called directly on the AbsDecoder class, as this class is intended to be subclassed.

######### Examples

>>> decoder = MyDecoder()  # MyDecoder is a subclass of AbsDecoder
>>> input_tensor = torch.randn(32, 100, 256)  # Example input
>>> input_lengths = torch.tensor([100] * 32)  # Example lengths
>>> output, additional_info = decoder.forward(input_tensor, input_lengths)

NOTE

This method must be overridden in any subclass of AbsDecoder to provide the specific decoding logic.

abstract property num_spk

Abstract property that must be implemented by subclasses to return the number of speakers that the decoder can handle.

This property is essential for the functioning of the decoder, as it informs the model how many unique speakers it should expect during the decoding process.

• Returns: The number of speakers that the decoder can process.
• Return type: int
• Raises:NotImplementedError – If the property is not implemented in the subclass.

######### Examples

class MyDecoder(AbsDecoder): : @property def num_spk(self): <br/>

return 2 # This decoder can handle 2 speakers.