Precision Automated Phonetic Analysis of Speech Signals for Information Technology of Text-dependent Authentication of a Person by Voice
Oleg V. Bisikalo, Олеся Бойван, Nina Khairova, Oksana V. Kovtun, Viacheslav V. Kovtun · 2021
A model of the process of phonetic analysis of speech signals in the frequency and temporal spaces is highlighted in the article for the first time. The generalization of the spectral characteristics of the studied speech signals is formalized in the represented model as an optimization task of minimizing the functional of relative entropy in contrast to the existing models. The obtained mathematical apparatus made it possible to formulate metric for quantitative estimation of the quality of the phonetic analysis results and to propose an adaptive method of automated phonetic analysis with an integrated mechanism for counteracting the influence of Gaussian-type noise, found in the studied speech signal, on the final result. The adequacy and functionality of the proposed model and method have been proved empirically. The analysis of the experiments results also showed that it is possible to assess the suitability of the studied speech materials for the task of authenticating a person by voice or speech recognition, focusing on the value of the coefficient of variability, which is included in the metric proposed by the authors and determined for the studied database of phonograms with recordings of voiced syllables of speech. Also, the values of this coefficient determined for the studied phonemes can be used to estimate the degree of their vocalization.