Databases Reference
In-Depth Information
on speech coding have changed. We have looked at several algorithms that have been designed
to operate in this changed environment. For images, fractal coding provides a very different
way to look at the problem. Instead of using the physical structure of the system to generate
the source output, it uses a more abstract view to obtain an analysis/synthesis technique.
Further Reading
1.
For information about various aspects of speech processing,
Voice and Speech Processing,
by T. Parsons [
119
], is a very readable source.
2.
The classic tutorial on linear prediction is “Linear Prediction,” by J. Makhoul [
252
],
which appeared in the April 1975 issue of the
Proceedings of the IEEE
.
3.
For a thorough review of recent activity in speech compression, see “Speech Coding
Methods, Standards, and Applications,” by J.D. Gibson [
245
], which appeared in
IEEE
Circuits and Systems Magazine
.
4.
An excellent source for information about speech coders is
Digital Speech: Coding for
Low Bit Rate Communication Systems
, by A. Kondoz [
285
].
5.
An excellent description of the G.728 algorithm can be found in “A Low Delay CELP
Coder for the CCITT 16kb/s Speech Coding Standard,” by J.-H. Chen, R.V. Cox, Y.-C.
Lin, N. Jayant, and M.J. Melchner [
240
], in the June 1992 issue of the
IEEE Journal on
Selected Areas in Communications
.
6.
A good introduction to fractal image compression is
Fractal Image Compression: Theory
and Application
, Y. Fisher (ed.) [
250
].
7.
The October 1993 issue of the
Proceedings of the IEEE
contains a special section on
fractals with a tutorial on fractal image compression by A. Jacquin.
18.8 Projects and Problems
1.
Write a program for the detection of voiced and unvoiced segments using the AMDF
function. Test your algorithm on the
test.snd
sound file.
2.
The
testf.raw
file is a female voice saying the word
test
. Isolate 10 voiced and
unvoiced segments from the
testm.raw
file and the
testf.snd
file. (Try to pick the
same segments in the two files.) Compute the number of zero crossings in each segment
and compare your results for the two files.
3.
(a)
Select a voiced segment from the
testf.raw
file. Find the fourth-, sixth-, and
tenth-order LPC filters for this segment using the Levinson-Durbin algorithm.
(b)
Pick the corresponding segment from the
testf.snd
file. Find the fourth-, sixth-,
and tenth-order LPC filters for this segment using the Levinson-Durbin algorithm.
(c)
Compare the results of (a) and (b).
4.
Select a voiced segment from the
test.raw
file. Find the fourth-, sixth-, and tenth-
order LPC filters for this segment using the Levinson-Durbin algorithm. For each of the
filters, find the multipulse sequence that results in the closest approximation to the voiced
signal.