Automatic speech signal segmentation with chosen parametrization method

Abstract

This paper is dedicated to the problem of automatic segmentation of a speech signal into so-called phonetic segments, i.e. speech signal segment with homogeneous physical structure which can be ascribed with adequate phonetic mean. This is the second trend in segmentation, as opposed to speech signal segmentation into short fixed segments. A segmentation algorithm is presented. It is based on calculations of the phonetic function at speech, which makes it possible to find the boundaries of these phonetic segments. The usability of three different parametrization methods - based on the analysis of zero-crossings, spectral analysis and linear prediction coding - was analyzed. No significant differences were observed in the efficiency of investigated parameters.