Abstract
In this paper, we present a minimum cross entropy spectrum estimation method (MCEM) with the constraints of uncertain autocorrelations. Based on the MCE principle using a prior spectrum for speech data, we show the results of applying to the so-called Kan-Non system. MCEM has been proposed by Shore and Johnson, that is an extension of the maximum entropy spectral estimation method (MEM) by Burg. Subsequently the original MCEM has been extended to more practical and general directions. Here, we derive the computational algorithms for the continuous spectral estimation by MCEM with the constraints of uncertain autocorrelations. Further, we show the results of the numerical simulation of the proposed method.
