espnet2.asvspoof.loss.binary_loss.ASVSpoofBinaryLoss

source

class espnet2.asvspoof.loss.binary_loss.ASVSpoofBinaryLoss(weight: float = 1.0)

Bases: AbsASVSpoofLoss

Binary loss for ASV Spoofing.

This class implements the binary cross-entropy loss function for anti-spoofing tasks in speaker verification. The loss is computed based on the predicted probabilities of the model and the ground truth labels.

weight

A scalar value to weight the loss. Default is 1.0.

• Type: float

• Parameters:weight (float) – A scalar value to weight the loss during training. Default is 1.0.

forward(pred

torch.Tensor, label: torch.Tensor,

**

kwargs): Computes the binary cross-entropy loss based on the model predictions and the ground truth labels.

score(pred

torch.Tensor) -> torch.Tensor: Returns the predictions as is, which can be used for evaluation metrics.

######### Examples

>>> loss_fn = ASVSpoofBinaryLoss(weight=0.5)
>>> predictions = torch.tensor([[0.8, 0.2], [0.1, 0.9]])
>>> labels = torch.tensor([[1, 0], [0, 1]])
>>> loss = loss_fn(predictions, labels)
>>> print(loss.item())

####### NOTE The predictions should be the raw output from the model and should be of shape [Batch, 2], where the second dimension represents the two classes (genuine and spoof). The labels should be in the same shape.

• Raises:ValueError – If the shape of predictions does not match the shape of labels.

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

#

forward(pred

: Tensor, label: Tensor, **kwargs)

Compute the binary cross-entropy loss for ASV spoofing.

This method calculates the binary cross-entropy loss between the predicted probabilities and the ground truth labels. The predictions are first passed through a sigmoid function to obtain values between 0 and 1 before calculating the loss.

• Parameters:
  • pred (torch.Tensor) – Prediction probabilities of shape [Batch, 2], where each row corresponds to the predicted scores for the two classes (genuine and spoof).
  • label (torch.Tensor) – Ground truth labels of shape [Batch, 2], where each row contains the binary labels (0 for genuine, 1 for spoof) corresponding to the predictions.
• Returns: The computed binary cross-entropy loss as a scalar tensor.
• Return type: torch.Tensor

######### Examples

>>> import torch
>>> loss_fn = ASVSpoofBinaryLoss()
>>> pred = torch.tensor([[0.9, 0.1], [0.2, 0.8]])
>>> label = torch.tensor([[1, 0], [0, 1]])
>>> loss = loss_fn.forward(pred, label)
>>> print(loss)  # Output will be a tensor representing the loss

####### NOTE The input tensors pred and label should have the same shape, and the values in label should be binary (0 or 1).

• Raises:ValueError – If the shapes of pred and label do not match.

#

score(pred

: Tensor)

Calculate the score from the prediction tensor.

This method takes the prediction tensor and returns it directly as the score. The score can be interpreted as the raw output probabilities from the model.

• Parameters:pred (torch.Tensor) – The prediction probabilities from the model, expected shape [Batch, 2].
• Returns: The input prediction tensor, serving as the score.
• Return type: torch.Tensor

######### Examples

>>> import torch
>>> loss_fn = ASVSpoofBinaryLoss()
>>> pred = torch.tensor([[0.8, 0.2], [0.4, 0.6]])
>>> score = loss_fn.score(pred)
>>> print(score)
tensor([[0.8, 0.2],
        [0.4, 0.6]])

####### NOTE The returned score is the same as the input predictions without any further processing.