nnmnkwii.functions.modspec¶

nnmnkwii.functions.modspec(y, n=4096, norm=None)[source]¶

Modulation spectrum computation

Given an parameter trajectory (T x D), it computes modulation spectrum. Here we define modulation spectrum is power of discrete Fourier transform of parameter trajectory. See [R6] for example application.

[R6]	Takamichi, Shinnosuke, et al. “A postfilter to modify the modulation spectrum in HMM-based speech synthesis.” Acoustics, Speech and Signal Processing (ICASSP), 2014 IEEE International Conference on. IEEE, 2014.

Parameters:	n – DFT length norm – Normalization mode. See `numpy.fft.fft()`
Returns:	Modulation spectrum as `T x n//2 + 1` array.
Return type:	numpy.ndarray

See also

nnmnkwii.autograd.modspec()

Examples

>>> import numpy as np
>>> from nnmnkwii import functions as F
>>> generated = np.random.rand(10, 2)
>>> ms = F.modspec(generated, n=16)
>>> ms.shape
(9, 2)