espnet2.gan_tts.hifigan.loss.FeatureMatchLoss

source

class espnet2.gan_tts.hifigan.loss.FeatureMatchLoss(average_by_layers: bool = True, average_by_discriminators: bool = True, include_final_outputs: bool = False)

Bases: Module

Feature matching loss module.

This module calculates the feature matching loss used in generative adversarial networks (GANs) to ensure that the generated outputs have similar features to the ground truth outputs. The loss can be averaged across layers and discriminators, and it can optionally include the final outputs of the discriminators for loss calculation.

average_by_layers

Whether to average the loss by the number of layers.

• Type: bool

average_by_discriminators

Whether to average the loss by the number of discriminators.

• Type: bool

include_final_outputs

Whether to include the final output of each discriminator for loss calculation.

• Type: bool

• Parameters:

  • average_by_layers (bool) – Whether to average the loss by the number of layers.
  • average_by_discriminators (bool) – Whether to average the loss by the number of discriminators.
  • include_final_outputs (bool) – Whether to include the final output of each discriminator for loss calculation.

• Returns: Feature matching loss value.

• Return type: Tensor

Examples

>>> feature_match_loss = FeatureMatchLoss()
>>> feats_hat = [torch.randn(2, 80, 100), torch.randn(2, 80, 100)]
>>> feats = [torch.randn(2, 80, 100), torch.randn(2, 80, 100)]
>>> loss = feature_match_loss(feats_hat, feats)
>>> print(loss)

Initialize FeatureMatchLoss module.

• Parameters:
  • average_by_layers (bool) – Whether to average the loss by the number of layers.
  • average_by_discriminators (bool) – Whether to average the loss by the number of discriminators.
  • include_final_outputs (bool) – Whether to include the final output of each discriminator for loss calculation.

forward(feats_hat: List[List[Tensor]] | List[Tensor], feats: List[List[Tensor]] | List[Tensor]) → Tensor

Calculate generator adversarial loss.

• Parameters:outputs (Union *[*List *[*List *[*Tensor ] ] , List *[*Tensor ] , Tensor ]) – Discriminator outputs, list of discriminator outputs, or list of list of discriminator outputs.
• Returns: Generator adversarial loss value.
• Return type: Tensor