Speaker identification using features based on first order Bessel function expansion of speech

The similarity between voiced speech and the Bessel functions yields a compact representation of speech using the nonsinusoidal Bessel functions. This paper presents the results of speaker identification experiments using features obtained from the first order Bessel function expansion of speech frames. Identification results on two large data bases using features based on {J₁(t)} expansion showed scores ranging from 55% to 75%. The scores depended on the indices of the Bessel functions used in the feature representation