ipqmf#

diffsptk.IPQMF#: alias of PseudoQuadratureMirrorFilterBankSynthesis

class diffsptk.PseudoQuadratureMirrorFilterBankSynthesis(n_band: int, filter_order: int, alpha: float = 100, learnable: bool = False, **kwargs)[source]#

See this page for details.

Parameters:

n_bandint >= 1: The number of subbands, \(K\).
filter_orderint >= 2: The order of the filters, \(M\).
alphafloat > 0: The stopband attenuation in dB.
learnablebool: Whether to make the filter-bank coefficients learnable.
**kwargsadditional keyword arguments: The parameters to find optimal filter-bank coefficients.

References

[1]

T. Q. Nguyen, “Near-perfect-reconstruction pseudo-QMF banks,” IEEE Transactions on Signal Processing, vol. 42, no. 1, pp. 65-76, 1994.

[2]

F. Cruz-Roldan et al., “An efficient and simple method for designing prototype filters for cosine-modulated filter banks,” IEEE Signal Processing Letters, vol. 9, no. 1, pp. 29-31, 2002.

forward(y: Tensor, keepdim: bool = True) → Tensor[source]#

Reconstruct waveform from subband waveforms.

Parameters:

yTensor [shape=(B, K, T) or (K, T)]: The subband waveforms.
keepdimbool: If True, the output shape is (B, 1, T) instead of (B, T).

Returns:

outTensor [shape=(B, 1, T) or (B, T)]: The reconstructed waveform.

Examples

>>> x = torch.arange(0, 1, 0.25)
>>> x
tensor([0.0000, 0.2500, 0.5000, 0.7500, 1.0000])
>>> pqmf = diffsptk.PQMF(2, 10)
>>> ipqmf = diffsptk.IPQMF(2, 10)
>>> x2 = ipqmf(pmqf(x), keepdim=False)
>>> x2
tensor([[[8.1887e-04, 2.4754e-01, 5.0066e-01, 7.4732e-01, 9.9419e-01]]])