Efficient LPC Quantization Methods - Digital Speech: Coding for Low Bit Rate Communication Systems - page 143

and percentage of 2 dB outliers respectively. The percentage of 4 dB outliers is
shown in Table 5.17. For VQ bit allocation greater than 18 bits, the percentage
of 4 dB outliers is zero.
Informal listening tests also showed the superiority of the new method at
the expense of increased complexity.

Table 5.17 Percentage of 4 dB outliers
VQ bit allocation

15 bits 16 bits 17 bits 18 bits

g

0.0059 0.0059 0.0 0.0

f

0.0415 0.0119 0.0059 0.0

5.11 Summary

This chapter has presented the fundamental aspects of speech spectral repre-
sentation via linear prediction. Accurate spectral representation of speech is
crucial to the performance of low bit-rate speech coders, especially in sinu-
soidal coders where the simplified excitation model cannot compensate for
shortcomings in the LPC modelling. Various quantization techniques have
been investigated and have lead to the design of LSF quantization schemes
optimized for specific configurations. These schemes offer solutions to LPC
parameter quantization in the form of LSFs for several applications, with
varying degrees of performance at a given bit rate and implementation com-
plexity. A more fundamental approach to improving quantization by way
of anti-aliasing filtering has also been presented with increased quantization
performance.

Bibliography

[1] S. Saito and K. Nakata (1985) Fundamentals of Speech Signal Processing ,
Chapter 9. Academic Press
[2] F. Soong and B. H. Juang (1984) 'Line spectrum pairs and speech data
compression', in Proc. of Int. Conf. on Acoust., Speech and Signal Processing ,
pp. 1.10.1-4.
[3] L. Rabiner and R. Schafer (1978) Digital Processing of Speech Signals .
Englewood Cliffs, NJ: Prentice-Hall
[4] B. Atal, R. V. Cox, and P. Kroon (1989) 'Spectral quantization and
interpolation for CELP coders', in Proc. of Int. Conf. on Acoust., Speech and
Signal Processing , pp. 69-72.
[5] G. Kang and C. J. Francen (1984) 'Low bit rate speech encoders based on
line spectrum frequencies (LSF)', Tech. Rep. 8857 , U.S. Naval Lab. Report.