Analysis by Synthesis LPC Coding - Digital Speech: Coding for Low Bit Rate Communication Systems - page 254

with almost negligible distortion in the speech. The post-filtered speech is
characterized by its lack of background noise components (quiet room effect)
and increased smoothness for voiced speech. For lower-rate CELPs, this
enhancement to the subjective quality is particularly noticeable: the speech
sounds much cleaner and much more pleasant to listen to. As suggested
in [17], making the high-pass factor, µ ,adaptiveas

= | |

µ ε k 1 (7.120)

where is the modulus of the first reflection coefficient computed from the

quantized LP parameters and ε is a tuning factor with a typical value of 0.3,
improves the speech quality.

| k 1 |

7.4 Summary

Analysis by synthesis coding of speech in the form of MPLPC and CELP has
been very popular for the past couple of decades. At bit rates of 6 kb/s and
above they produce good performance and the various versions reported in
the literature differ mainly on the way the secondary (codebook) excitation
is generated or represented. In early days, random Gaussian numbers were
used to populate the codebooks, but they were complex to store and search
and did not produce the best quality. The use of vector sum excitation
improved the situation both in terms of the cost of implementation and the
overall speech quality. However the most successful CELP coders have been
produced after the invention of algebraic codebooks. ACELPs are currently
being used in many international standards.
AlthoughACELPs have been very dominant at bit rates of 6 kb/s and above,
they rely heavily on objective measures (although perceptual weighting is
used) and as the bit rate is lowered their quality deteriorates rapidly. It would
therefore be very difficult to produce a toll-quality 4 kb/s CELP coder unless
significant modifications are made to the basic structure described above.

Bibliography

[1] B. Atal and M. Schroeder (1970) 'Adaptive predictive coding of speech
signals', in Bell Sys. Technical Journal , pp. 1973-87. October 1970.
[2] M. Schroeder and B. Atal (1979) 'Predictive coding of speech signals
and subjective error criteria', in IEEE Trans. on Acoust., Speech and Signal
Processing , 27:247-54.
[3] R. Zelinski and P. Noll (1977) 'Adaptive transform coding of speech sig-
nals', in IEEETrans. onAcoust., Speech and Signal Processing , 25(4):299-309.