ИММУНОЛОГИЧЕСКАЯ МОДЕЛЬ ТЕКСТОНЕЗАВИСИМОЙ ГОЛОСОВОЙ ИДЕНТИФИКАЦИИ ЛИЧНОСТИ

Abstract

An immunological model of clonal selection with positive selection based on the principlesof mass-parallel data processing used in artificial immune systems, is proposed. The model isdesigned for text-independent identification of a person by voice. In contrast to known passwordbasedvoice identification systems, the proposed model implements decentralized recognition ofvoice data by matching it with detectors that simulate immunocompetent cells of the immune system.The initial voice features are generated in a linear speech predictor and are represented bycepstral coefficients. The sequence of cepstral coefficients is further divided into equal time sections- morphemes, which are abstract linguistic units that unify phonemes. Morphemes carry theindividual coloring of consecutive temporal segments of speech reproduced by the voice, allowingthem to be used productively as voice identifiers. The matching of voice morphemes with detectorsis carried out according to the principle of positive selection based on the Euclidean proximitymeasure. The model's "friend-or-foe" identification decision making is implemented on the basis ofa statistical approach in terms of the frequency of detector response. The proposed model implementsthe identification of the speaker's personality at the rate of receipt of his voice data. At thesame time, personality identification is invariant to the language, volume and content of speech.The advantage of the model is complete protection against replay attacks. The effective realizationof the model, the accuracy and speed of identification are due to the possibility of organizing highspeedanalysis of large volumes of voice data, which in the long term corresponds to the pace ofdevelopment and application of high-performance computing systems.