The basic elements for cognitive model of speech perception mechanism on the base of multi-agent recursive intellect

In this paper, the generalized architecture used in almost all modern systems of automatic speech recognition is analyzed. The necessity of developing a fundamentally new approach to solving speech recognition problems is outlined. A formal description of the structure of the speech perception act is proposed for use as a general theoretical basis in the development of universal automatic speech recognition systems that are highly effective in conditions of high noise and “cocktail party” situations. The general structural dynamics of the speech recognition process has been developed, which allows to take into account the linguistic and extra-linguistic aspects of a speech message. The concept of an articulation event as a minimal basic pattern of sound image recognition has been proposed. The recognition process is structured based on the functional determinants of the situation. The need to analyze the numerous sources of information accompanying the sound message, the rejection of the search for an invariant here is of fundamental nature. Multi-agent systems were chosen as the formal means for implementation. Multi-agent approach allows to differentiate and analyze sounds of different nature. This makes the proposed model unique and gives it advantages in the so-called “cocktail party” situation, as well as in tasks where the noise level is extremely high.