Method of Automated Transcribing of Speech Signals for Information Technology of Text-Dependent Authentication of a Person by Voice

A brand new method of automated transcribing of speech signals of the selected language is represented in the article. In this method, in contrast to the existing ones, the result of transcribing is obtained as a solution of the task of optimization of the functional of relative entropy for each of the sequences of non-intersecting frames, which generalize the samples counts of the sampled speech signal. The method provides an option to improve the quality of transcribing due to personalized pre-setting of the bank of whitening filters on estimating the power spectral density in the target frequency domains of the characteristic phonemes of the selected language. Also on the basis of the proposed method in the first approximation the task of classification of language units of higher levels, in particular, morphemes and lexemes, is set. The adequacy of the proposed mathematical apparatus is proved empirically.