M-ary predictive coding: a nonlinear model for speech

Speech Coding is pivotal in the ability of networks to support multimedia services. The technique currently used for speech coding is Linear Prediction. It models the throat as an all-pole filter i.e. using a linear difference equation. However, the physical nature of the throat is itself a clue to its nonlinear nature. Developing a nonlinear model is difficult as in the solution of nonlinear equations and the verification of nonlinear schemes. In this paper, two nonlinear models– Quadratic Predictive Coding and M-ary Predictive Coding-have been proposed. The equations for the models were developed and the coefficients solved for. MATLAB was used for implementation and testing. This paper gives the equations and preliminary results. The proposed enhancements to MPC are also discussed in brief. 1. AIM The objective of this endeavour was to i. Investigate the two nonlinear models proposed by the author for more effective speech coding. ii. Development of one of the two models – MPC (M-ary Predictive Coding) – as a complete coding scheme. 2. INTRODUCTION Speech is a very special signal for a variety of reasons. The most preliminary of these is the fact that speech is a non-stationary signal. This makes speech a rather tough signal to analyze and model. The second reason, is that factors like the intelligibility, the coherence and other such 'human' characteristics play a vital role in speech analysis as against statistical parameters like Mean Square Error. The third reason is from a communications point of view. The number of discrete values required to describe one second of speech amounts to 8000 (at the minimum). Due to concerns of bandwidth, compression is desirable. In the technique of linear prediction, the vocal tract is modeled as an all-pole filter[1]. That is, the transfer function of the vocal throat is written as i=1 In terms of a difference equation, p y[n]= Σa i *y[n-i] i=1