# ------------------------------------------------------------------------ #
# Copyright 2022 SPTK Working Group #
# #
# Licensed under the Apache License, Version 2.0 (the "License"); #
# you may not use this file except in compliance with the License. #
# You may obtain a copy of the License at #
# #
# http://www.apache.org/licenses/LICENSE-2.0 #
# #
# Unless required by applicable law or agreed to in writing, software #
# distributed under the License is distributed on an "AS IS" BASIS, #
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #
# See the License for the specific language governing permissions and #
# limitations under the License. #
# ------------------------------------------------------------------------ #
import torch
from ..typing import Precomputed
from ..utils.private import check_size, get_gamma, get_values
from .base import BaseFunctionalModule
[docs]
class GeneralizedCepstrumGainNormalization(BaseFunctionalModule):
"""See `this page <https://sp-nitech.github.io/sptk/latest/main/gnorm.html>`_
for details.
Parameters
----------
cep_order : int >= 0
The order of the cepstrum, :math:`M`.
gamma : float in [-1, 1]
The gamma parameter, :math:`\\gamma`.
c : int >= 1 or None
The number of filter stages.
References
----------
.. [1] T. Kobayashi et al., "Spectral analysis using generalized cepstrum," *IEEE
Transactions on Acoustics, Speech, and Signal Processing*, vol. 32, no. 5,
pp. 1087-1089, 1984.
"""
def __init__(self, cep_order: int, gamma: float = 0, c: int | None = None) -> None:
super().__init__()
self.in_dim = cep_order + 1
self.values = self._precompute(*get_values(locals()))
[docs]
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""Perform cepstrum gain normalization.
Parameters
----------
x : Tensor [shape=(..., M+1)]
The generalized cepstrum.
Returns
-------
out : Tensor [shape=(..., M+1)]
The normalized generalized cepstrum.
Examples
--------
>>> x = diffsptk.ramp(1, 4)
>>> gnorm = diffsptk.GeneralizedCepstrumGainNormalization(3, c=2)
>>> y = gnorm(x)
>>> y
tensor([2.2500, 1.3333, 2.0000, 2.6667])
"""
check_size(x.size(-1), self.in_dim, "dimension of cepstrum")
return self._forward(x, *self.values)
@staticmethod
def _func(x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
values = GeneralizedCepstrumGainNormalization._precompute(
x.size(-1) - 1, *args, **kwargs
)
return GeneralizedCepstrumGainNormalization._forward(x, *values)
@staticmethod
def _takes_input_size() -> bool:
return True
@staticmethod
def _check(cep_order: int, gamma: float, c: int | None) -> None:
if cep_order < 0:
raise ValueError("cep_order must be non-negative.")
if 1 < abs(gamma):
raise ValueError("gamma must be in [-1, 1].")
if c is not None and c < 1:
raise ValueError("c must be greater than or equal to 1.")
@staticmethod
def _precompute(cep_order: int, gamma: float, c: int | None = None) -> Precomputed:
GeneralizedCepstrumGainNormalization._check(cep_order, gamma, c)
return (get_gamma(gamma, c),)
@staticmethod
def _forward(x: torch.Tensor, gamma: float) -> torch.Tensor:
x0, x1 = torch.split(x, [1, x.size(-1) - 1], dim=-1)
if gamma == 0:
K = torch.exp(x0)
y = x1
else:
z = 1 + gamma * x0
K = torch.pow(z, 1 / gamma)
y = x1 / z
y = torch.cat((K, y), dim=-1)
return y
