espnet2.enh.loss.criterions.time_domain.SNRLoss

source

class espnet2.enh.loss.criterions.time_domain.SNRLoss(eps=1.1920928955078125e-07, name=None, only_for_test=False, is_noise_loss=False, is_dereverb_loss=False)

Bases: TimeDomainLoss

SNR (Signal-to-Noise Ratio) loss for time-domain enhancement.

This loss computes the negative signal-to-noise ratio (SNR) between the reference signal and the estimated signal. The SNR is calculated in decibels, where a higher SNR indicates better performance.

• Parameters:
  • eps (float) – A small constant added to the denominator to avoid division by zero (default: machine epsilon).
  • name (str , optional) – Name of the loss function (default: “snr_loss”).
  • only_for_test (bool , optional) – If True, this loss is only used during testing (default: False).
  • is_noise_loss (bool , optional) – If True, this loss is used for noise related tasks (default: False).
  • is_dereverb_loss (bool , optional) – If True, this loss is used for dereverberation tasks (default: False).
• Returns: The computed SNR loss, with shape (Batch,).
• Return type: torch.Tensor

####### Examples

>>> snr_loss = SNRLoss()
>>> reference = torch.tensor([[1.0, 2.0, 3.0], [1.0, 2.0, 3.0]])
>>> estimate = torch.tensor([[1.0, 1.0, 1.0], [2.0, 2.0, 2.0]])
>>> loss = snr_loss(reference, estimate)
>>> print(loss)
tensor([-3.0103, -3.0103])  # Example output, will vary based on input.

• Raises:AssertionError – If the shapes of the reference and estimated tensors do not match.

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

forward(ref: Tensor, inf: Tensor) → Tensor

Signal-to-Noise Ratio (SNR) Loss.

This class computes the Signal-to-Noise Ratio (SNR) loss, which is defined as the difference in decibels between the signal and the noise. The SNR is a critical metric in evaluating the quality of audio signals, especially in speech enhancement tasks.

eps

A small value to prevent division by zero.

• Type: float

• Parameters:

  • eps (float) – A small value to avoid numerical instability (default: machine epsilon).
  • name (str , optional) – Name of the loss function (default: “snr_loss”).
  • only_for_test (bool , optional) – Flag indicating if the loss is only for testing (default: False).
  • is_noise_loss (bool , optional) – Flag indicating if this loss is for noise estimation (default: False).
  • is_dereverb_loss (bool , optional) – Flag indicating if this loss is for dereverberation (default: False).

• Raises:ValueError – If both is_noise_loss and is_dereverb_loss are True.

####### Examples

>>> import torch
>>> snr_loss = SNRLoss()
>>> reference = torch.tensor([[1.0, 2.0], [3.0, 4.0]])
>>> estimated = torch.tensor([[1.0, 2.5], [2.5, 4.5]])
>>> loss = snr_loss(reference, estimated)
>>> print(loss)
tensor([-2.0000, -2.0000])

Notes

The returned loss value is negative, representing the loss to be minimized during training.