Digital Signal Processing Reference
In-Depth Information
a
ML-MVQ
ˆ
m
+
Σ
ω
0
ˆ
r
+
+
a
ˆ
Mel to
Linear
Σ
−
ˆ
r
+
Σ
+
First Order
Predictor
ˆ
p
P-MVQ
Mean Square Error
Minimization
Perceptual
Weighting
Figure 9.30
Block diagram of SP-MVQ
frequencies, taking into account the perceptual preferences of the human
auditory system. The fixed dimension spectral vector,
z
,isdecomposedinto
ˆ
a predicted vector,
z
p
, and a prediction residual vector,
ˆ
z
r
, as follows:
ˆ
ˆ
z
= ˆ
z
p
+ ˆ
z
r
(9.62)
where the predicted vector,
ˆ
z
p
, is obtained using a first-order autoregressive
method, given by,
z
p
=
ˆ
z
m
+ ˆ
ˆ
z
− ˆ
z
m
(9.63)
−
1
where
z
m
is the mean vector, and
denotes a diagonal matrix of prediction coefficients. The prediction residual,
ˆ
z
ˆ
1
is the most recently quantized
z
,
ˆ
ˆ
−
z
r
is quantized using a typical vector quantizer such as MSVQ [52]. The
quantization becomes memoryless Mel-scale-based vector quantization (ML-
MVQ) if all the prediction coefficients are zero, and autoregressive predictive
MVQ (P-MVQ) otherwise. The predictive scheme is effective in stationary
regions, and may increase spectral distortion at the transitions; therefore, a
switching scheme is introduced to switch between P-MVQ andML-MVQ. The
decision between P-MVQ and ML-MVQ is made using AbS techniques and
Search WWH ::
Custom Search