The implementation of linear prediction techniques for all but trivial examples requires digital computation. In nearly all of the material presented thus far, an implicit assumption was made that the computation occurred without error. For example, the norm squaresα_m andβ_m of Chapter 3 are theoretically non-negative, yet when they are evaluated using the computer programs presented they may yield negative results due to computer errors. Numerical results are a function of the algorithms used, the form of computer arithmetic (e.g., floating or fixed point), and the computer word length.

Unable to display preview. Download preview PDF.

Authors and Affiliations

1. Speech Communications Research Laboratory, Inc., 93109, Santa Barbara, California, USA

   John D. Markel & Augustine H. Gray Jr.

2. Department of Electrical Engineering and Computer Science, University of California Santa Barbara, 93106, California, USA

   Augustine H. Gray Jr.

Reprints and permissions

© 1976 Springer-Verlag Berlin Heidelberg

Policies and ethics