espnet2.enh.layers.beamformer.get_gev_vector

source

espnet2.enh.layers.beamformer.get_gev_vector(psd_noise: Tensor | ComplexTensor, psd_speech: Tensor | ComplexTensor, mode='power', reference_vector: int | Tensor = 0, iterations: int = 3, diagonal_loading: bool = True, diag_eps: float = 1e-07, eps: float = 1e-08) → Tensor | ComplexTensor

Return the generalized eigenvalue (GEV) beamformer vector.

The GEV beamformer is defined by the equation: : psd_speech @ h = lambda * psd_noise @ h

Reference: : Blind acoustic beamforming based on generalized eigenvalue decomposition; E. Warsitz and R. Haeb-Umbach, 2007.

• Parameters:

  • psd_noise (torch.complex64/ComplexTensor) – Noise covariance matrix of shape (…, F, C, C).

  • psd_speech (torch.complex64/ComplexTensor) – Speech covariance matrix of shape (…, F, C, C).

  • mode (str) –

    One of (“power”, “evd”).

    • “power”: Power method for calculating the eigenvector.
    • “evd”: Eigenvalue decomposition (only for torch builtin

    complex tensors).

  • reference_vector (torch.Tensor or int) – Reference vector of shape (…, C) or a scalar.

  • iterations (int) – Number of iterations for the power method.

  • diagonal_loading (bool) – Whether to add a small term to the diagonal of psd_noise to prevent singularity.

  • diag_eps (float) – Regularization parameter for diagonal loading.

  • eps (float) – Small constant to avoid division by zero.

• Returns: GEV beamformer vector of shape (…, F, C).
• Return type: beamform_vector (torch.complex64/ComplexTensor)

Examples

>>> psd_n = torch.rand(10, 8, 4, 4, dtype=torch.complex64)
>>> psd_s = torch.rand(10, 8, 4, 4, dtype=torch.complex64)
>>> vector = get_gev_vector(psd_n, psd_s, mode="power")
>>> print(vector.shape)
torch.Size([10, 8, 4])