Speech Analysis with Bessel Functions

Abstract

A signal may be expressed as a I inear combination of other functions, cal led the basis set. This is essentially a model of the waveform or signal. Infinitely many choices are possible for the basis set, with the most common choice being the set of trigonometric functions. This study involves the use of Bessel functions of the first kind as the basis set. The original goal of the research was to develop an automatic speaker recognition scheme, based upon the Fourier Bessel series. But the difficulty of collecting a high quality data base and certain hardware deficiencies precluded the completion of the original goal. Also, it was found that the theoretical foundation for the use of the Fourier-Bessel series for signal analysis was practically nonexistent. For these reasons, the study was confined to general purpose speech analysis and to investigation of the computational algorithms required.