Digital Signal Processing Reference
In-Depth Information
Fig. 11.1 Example of variations of log-spectral coefficients generated by applying a weight to the
(a) zeroth, (b) first, and (c) fourth cepstral coefficients
time-varying noise included in the speech duration would reflect variations of the
estimated basis model. The variational models are generated by selectively applying
weights to each component of the mean vector of the basis model in the cepstral
domain. Here, we propose a novel algorithm to generate a collection of variational
noisemodelsasfollows:
Step 1 - Basis Model Estimation
A basis noise model is obtained from silent segments within the input speech, which
generally exists at the beginning and end parts of an utterance. The model is
estimated as a Gaussian pdf (
2
) in the cepstral domain.
m
,
s
Step 2 - Variational Component Determination
The
V
largest components {
v
1
,
v
2
,
2
are selected.
These are named
variational components,
which are considered highly variable
components and are size-ordered ranked as follows:
...
,
v
V
} in the variance vector
s
s
v
1
s
v
2
s
v
V
(11.2)
Search WWH ::
Custom Search