cdist¶

class diffsptk.CepstralDistance(full=False, reduction='mean')[source]¶

See this page for details.

Parameters

fullbool [scalar]: If true, include the constant term in the distance calculation.
reduction[‘none’, ‘mean’, ‘batchmean’, ‘sum’]: Reduction type.

forward(c1, c2)[source]¶

Calculate cepstral distance between two inputs.

Parameters

c1Tensor [shape=(…, M+1)]: Input cepstral coefficients.
c2Tensor [shape=(…, M+1)]: Target cepstral coefficients.

Returns

distTensor [shape=(…,) or scalar]: Cepstral distance.

Examples

>>> c1 = torch.randn(2, 3)
tensor([[ 0.4296,  1.6517, -0.6022],
        [-1.0464, -0.6088, -0.9274]])
>>> c2 = torch.randn(2, 3)
tensor([[ 1.6441, -0.6962, -0.2524],
        [ 0.9344,  0.3965,  1.1494]])
>>> cdist = diffsptk.CepstralDistance()
>>> dist = cdist(c1,c2)
>>> dist
tensor(1.6551)