espnet2.enh.layers.wpe.get_correlations

source

espnet2.enh.layers.wpe.get_correlations(Y: Tensor | ComplexTensor, inverse_power: Tensor, taps, delay) → Tuple[Tensor | ComplexTensor, Tensor | ComplexTensor]

Calculates weighted correlations of a window of length taps.

This function computes the correlation matrix and correlation vector for a given complex-valued Short-Time Fourier Transform (STFT) signal using a weighted approach based on the provided inverse power and correlation parameters.

espnet2.enh.layers.wpe.Y

Complex-valued STFT signal with shape (F, C, T).

• Type: Union[torch.Tensor, ComplexTensor]

espnet2.enh.layers.wpe.inverse_power

Weighting factor with shape (F, T).

• Type: torch.Tensor

espnet2.enh.layers.wpe.taps

Length of the correlation window.

• Type: int

espnet2.enh.layers.wpe.delay

Delay for the weighting factor.

• Type: int

• Parameters:

  • Y – Union[torch.Tensor, ComplexTensor] Complex-valued STFT signal with shape (F, C, T).
  • inverse_power – torch.Tensor Weighting factor with shape (F, T).
  • taps – int Length of correlation window.
  • delay – int Delay for the weighting factor.

• Returns:

  • Correlation matrix of shape (F, taps*C, taps*C).
  • Correlation vector of shape (F, taps, C, C).

• Return type: Tuple[Union[torch.Tensor, ComplexTensor], Union[torch.Tensor, ComplexTensor]]

• Raises:

  • AssertionError – If the dimensions of inverse_power do not match with
  • the dimensions of Y. –

Examples

>>> Y = torch.randn(4, 2, 10, dtype=torch.complex64)  # (F, C, T)
>>> inverse_power = torch.randn(4, 10)  # (F, T)
>>> taps = 5
>>> delay = 2
>>> correlation_matrix, correlation_vector = get_correlations(Y,
... inverse_power, taps, delay)
>>> print(correlation_matrix.shape)  # (4, 10, 10)
>>> print(correlation_vector.shape)   # (4, 5, 2, 2)

NOTE

This function assumes that the input tensors are properly shaped and the operations will be performed in a batch manner.