Analytical approximation of speech power spectral density in Serbian language

Estimation of speech power spectral density has special importance in speech analysis, synthesis and recognition. It is of practical interest to derive an adequate analytic function that describes speech power spectral density, like the one standardized by CCITT convention. In this paper, the analytical approximation of experimentally obtained speech power spectral density in Serbian language is performed by applying mean-square error minimization method. This method is applied in two different ways, and influence of some parameters in minimization process is investigated. The developed method is also applied for analytic approximation of power spectral density of English and Russian.