Digital Speech: Coding for Low Bit Rate Communication Systems, 2nd Edition

5

Efficient LPC Quantization Methods

5.1 Introduction

Linear predictive coding is a very powerful analysis technique and is used in many speech processing systems. In speech coding and synthesis systems, the analysis techniques for obtaining the LP coefficients (LPC), e.g. autocorrelation, covariance, lattice, and the quantization of the LPC are very important aspects of LPC analysis as minimization of coding capacity is the ultimate aim in these applications. The main objective of the quantization procedure is to code the LPC with as few bits as possible without introducing audible spectral distortion. Whilst perfect reconstruction is not possible, subjective transparency is achievable. Quantization of the LPC is usually performed by transforming the LPC to other forms which enables predictive coding and allows an easy filter stability check. The most popular LPC transformation is the use of Line Spectrum Pairs (LSP), related to the Line Spectral Frequency (LSF) representation of the LPC [1, 2]. In this chapter, the LSF representation of the LPC will be described, followed by various LPC quantization schemes using LSF transformation.

5.2 Alternative Representation of LPC

As was shown in Chapter 4, the LPC filter is given by

where p is the order of LPC filter.

The α_i coefficients are the direct form of LPC. The filter H(z) is stable if it is minimum phase, i.e. all the roots of ...