A.S. Kolokolov, V.M. Krol, A.Yu. Mestcherakov, I.A. Lubinski, V.P. Yachno

On the basis of the analysis of the data physiologists and psychophysics the possible hearing mechanisms providing stability of perception of speech in the presence of frequency distortions and background noise were revealed. At the heart of them three kinds of a bandpass filtering of an acoustical spectrum are used. The first two types of a bandpass filtering are realized by means of lateral inhibition processes with symmetric and asymmetrical neuron inhibition connections. Because of this filtering local maxima and sharp slopes are emphasized and brought to light in a spectral pattern of a sound. It is supposed that these selected spectral cues contain the important information for phoneme recognition. The third type of a bandpass filtering is realized on the basis of effect of the delayed inhibition and allows to detect local no uniformities of a speech spectral pattern in time. It is supposed that these detected hearing system sharp spectrum changes are used for the segmentation of a speech signal on the consecutive quasistationary segments representing single phonemes of a speech signal, and also for speech events duration estimation. It is well known that duration plays the important role in allocation in speech of stressed vowels, and also in division of explosive and fricative consonants. The study results can be used by working out the methods of signal processing in frequency domain as applied to problems of front end processing and analysis of speech and acoustic signals.