Digital Signal Processing Reference
In-Depth Information
80.0
(a)
60.0
40.0
(b)
20.0
0.0
0
5
10
15
20
Number of pulses / 32 samples
Figure 7.22
Comparison between (a) independent and (b) combinational pulse-
position coding
the first pulse magnitude and coded using fewer quantization levels (typi-
cally three bits) [6]. This assumes that the first pulse usually has the largest
magnitude and, if independent pulse position coding is used, this can be
advantageous. Otherwise the largest magnitude pulse may be used to limit
the pulses within
±
1.
7.3.4 CodebookExcitation
The vectors contained in the excitation codebook form a very important part
of the CELP coding algorithm. They serve two main purposes:
•
They provide the start-up information to the LTP memory, including
any sudden changes in the speech not adequately tracked by the pitch
prediction.
•
They supply the 'filling-in' information that the pitch predictor has omitted.
This is especially the case during unvoiced regions.
Thus, how the codebook of a CELP is populated and the method by which
the optimum vector is computed are very important issues as indicated by
the many publications on this subject [20-22]. Another related issue is the
computational cost and storage of the codebook search procedure. The search
processforthebestvectorinCELPcanbebrokendownintofourstages.
Search WWH ::
Custom Search