Digital Signal Processing Reference
In-Depth Information
The codebook
C
can be extended to the right by constructing it in the following
way. The first time, the
N
−
1 available samples
y
(
−
N
+1)
···
y
(
−
1) are used and
are supplemented by
y
(
−
N
+1); the second time around,
N
−
2 samples
y
(
−
N
+
2)
N
+3), etc. We stop at index
Q
min
. This artificially solves the problem of the size of the sub-frame being too high
for the fundamental period of a female speaker. Choosing
Q
min
= 20 is enough, for
example. We take:
···
y
(
−
1) are used, supplemented by
y
(
−
N
+2)
y
(
−
Q
max
=
Q
min
+ 127
if we decide to code the index on seven bits.
6.3.4.
Conclusion
We have provided here only the principles based on which CELP speech coders
are built. To obtain a reconstructed signal at a suitable quality, it is necessary to govern
precisely all the parameters in play and to add a number of tricks. To get an idea of
these, we can consult the paper [SAL 98].
We provide here information transmitted in the network in the case of the 8 kb/s
G.729 coder. The
P
=10coefficients of the synthesis filter 1
/A
(
z
) are updated for
every 10 ms (80 sample analysis frame). The LSP coding is realized with 18 bits.
The synthesis filter input is the sum of two signals updated for every 5 ms. The first
corresponds to the long-term predictor. The parameter
Q
and the gain are coded on
7 and 3 bits, respectively. The second comes from a vector quantization codebook.
The gain is coded on 4 bits and the index on 13+4 bits, conforming to what is indicated
in Table 6.1. Every 10 ms, 18+2
×
(3+7+4+17)=80bits must be transmitted,
requiring a bit rate of 8 kb/s.
Impulses Amplitude
Positions
Bits
0
±
1
0, 5, 10, 15, 20, 25, 30, 35 1+3
1
±
1
1, 6, 11, 16, 21, 26, 31, 36 1+3
2
±
1
2, 7, 12, 17, 22, 27, 32, 37 1+3
3
±
1
3, 8, 13, 18, 23, 28, 33, 38 1+4
4, 9, 14, 19, 24, 29, 34, 39
Tab l e 6 . 1 .
Every 40 samples of the G.729 coder select 4 “impulses”
for which the possible positions are indicated
We can also provide a few pieces of information on the 3GPP AMR-WB coder
(
Adaptive Multi-Rate Wideband
ITU-T G.722.2) [BES 02] which is the first coder
adapted for fixed and mobile networks. It removes the requirement for transcoding.
As its name indicates, several bit rates are possible. These vary between 6.6 and
