espnet2.spk.loss.abs_loss.AbsLoss

source

class espnet2.spk.loss.abs_loss.AbsLoss(nout: int, **kwargs)

Bases: Module

Abstract base class for loss functions in the ESPnet2 framework.

This class serves as a template for creating custom loss functions. It inherits from PyTorch’s nn.Module and requires the implementation of the forward method in derived classes.

nout

The number of output classes for the loss function.

• Type: int

• Parameters:

  • nout (int) – The number of output units for the loss function.
  • **kwargs – Additional keyword arguments for further customization.

forward(x

torch.Tensor, label=None) -> torch.Tensor: Computes the loss given the input tensor and the corresponding labels.

• Raises:
  • NotImplementedError – If the forward method is not implemented in a
  • derived class. –

####### Examples

class CustomLoss(AbsLoss): : def forward(self, x: torch.Tensor, label=None) -> torch.Tensor: : # Custom loss calculation here return loss_value

loss = CustomLoss(nout=10) output = loss(torch.randn(5, 10), label=torch.randint(0, 10, (5,)))

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

abstract forward(x: Tensor, label=None) → Tensor

Computes the forward pass of the loss function.

This method is an abstract method that must be implemented by any subclass of AbsLoss. It takes an input tensor and an optional label tensor to compute the loss.

• Parameters:
  • x (torch.Tensor) – The input tensor for which the loss is calculated.
  • label (optional) – The ground truth labels for the input tensor. Default is None.
• Returns: The computed loss value as a tensor.
• Return type: torch.Tensor
• Raises:NotImplementedError – If the method is not implemented in a subclass.

####### Examples

Example of a subclass implementation

class MyLoss(AbsLoss):

def forward(self, x, label=None): : return torch.mean(x) # Replace with actual loss computation

loss = MyLoss(nout=10) input_tensor = torch.randn(5, 10) output = loss.forward(input_tensor, label=torch.tensor([1, 0, 1, 0, 1]))