Digital Signal Processing Reference
In-Depth Information
Fig. 11.2 Mean parameters of variational models in log-spectral domain generated by the
proposed model composition methods. Four-digit symbol of each plot indicates a combination
of perturbation factors (i.e.,
a
,0,or
þ a
) for the selected four variational components
Step 3 - Model Composition by Mean Perturbation
A variation of the mean vector is generated by selectively applying the
perturbation
factor f
p
to the determined variational components of the cepstral coefficients
v
1
to
v
V
as follows:
m
i
ð
1
þ
f
p
Þ
if
i
2f
v
1
;
v
2
; ...;
v
V
g
m
i
¼
;
(11.3)
m
i
otherwise
where
f
p
¼a
is a small positive value which we determine
heuristically. The obtained model collection {
l¼
,0,or
þ a
and the
a
2
)} consists of a total
3
V
number of generated variational models as a result of combinations of the 3-
type gains of the
V
variational components.
In this study, we employed four variational components for the proposed model
compositionmethod. Figure
11.2
demonstrates several representative variational noise
models (i.e., mean parameters in the log-spectral domain) obtained by the proposed
model composition algorithm, showing various types of spectral patterns generated by
_
,
(
m
s
Search WWH ::
Custom Search