medfilt#

class diffsptk.MedianFilter(filter_length: int, across_features: bool = False, magic_number: float | None = None)[source]#

See this page for details.

Parameters:

filter_lengthint > 0: The length of the median filter, \(L\).
across_featuresbool: If True, apply the filter across the feature dimension.
magic_numberfloat or None: The magic number representing unvoiced frames.

forward(x: Tensor) → Tensor[source]#

Apply median filtering to the input sequence.

Parameters:

xTensor [shape=(B, N, D) or (N, D) or (N,)]: The input sequence.

Returns:

outTensor [shape=(B, N, D) or (B, N) or (N, D) or (N,)]: The filtered sequence.

Examples

>>> import torch
>>> import diffsptk
>>> medfilt = diffsptk.MedianFilter(3)
>>> x = torch.tensor([0, 2, -2, 7, 4, 8]).float()
>>> y = medfilt(x)
>>> y
tensor([1., 0., 2., 4., 7., 6.])

diffsptk.functional.medfilt(x: Tensor, filter_length: int = 3, across_features: bool = False, magic_number: float | None = None) → Tensor[source]#

Apply median filtering to the input sequence.

Parameters:

xTensor [shape=(B, N, D) or (N, D) or (N,)]: The input sequence.
filter_lengthint > 0: The length of the median filter, \(L\).
across_featuresbool: If True, apply the filter across the feature dimension.
magic_numberfloat or None: The magic number representing unvoiced frames.

Returns:

outTensor [shape=(B, N, D) or (B, N) or (N, D) or (N,)]: The filtered sequence.