The Concept of Automated Phonetic Analysis of a Speech with Asymptotic Adaptation to the Specifics of Phonation of Language Units

Abstract

A new concept of automated phonetic analysis of a speech with asymptotic adaptation to the specifics of phonation of language units is proposed. Unlike analogues, the concept is formalized in the paradigm of information theory with the criterion of analysis of the speech signal based on its relative entropy. The obtained mathematical apparatus allows for analysing the studied process both in statics and dynamics. The quality of the analysis is characterized by errors in the first (incorrect recognition of the language unit) and the second (variability of phonation of the language unit within the corresponding cluster) kind. The derived result is an analytically substantiated possibility of estimating the phonetic saturation of speech. The formulated concept can also be used to assess the degree of information saturation of the cluster of the language unit. Such an assessment is important for the task of creating universal background models, which are an essential element of current recognition systems for both language and speaker.