Automated definition of phonetically homogeneous sections of words in a natural language based on multiparameter optimization

Abstract

An approach to the automated splitting of words into phonetically homogeneous parts is proposed under which the boundaries of the parts are defined as a result of solving a multiparameter optimization problem. The approach is assumed to ensure the maximum difference in the phonetic material between the adjacent parts and the maximum similarity within the parts. The accepted measure of similarity and difference is based on the correlation between the columns of the parametric portrait matrix of the word generated as a result of a time-spectral conversion of an audio recording of the word. To obtain a numerical solution of the problem, an algorithm is proposed which is a modification of a dynamic programming technique. The experimental results are presented with several words from the Russian language taken as examples to confirm the legitimacy of the assumptions made and viability of the algorithms proposed.