A mathematical model for the determination of speech signal parameters with the aid of instant memory

In this paper the concept is presented of a system for the determination of basic parameters of speech signals, including formants and their transients. These parameters are described in stages by means of characteristic functions of the sound features. Values of the characteristic functions for the j-th stage of the sound features are arguments for the characteristic functions of other features of the (j+1)-st stage. If the values of these functions equal unity, they are automatically recorded in the memory. The length of the time of storage of these values is determined by physical properties of the sound. Physical properties at the (j+1)-st stage are described by means of relations that establish the sequence of appearance of the sound features as a function of time at the output of the j-th stage.
The system enables the analysis of sounds in real time, as well as the automatic segmentation of sound as a function of time. The accuracy of the identification of speech does not depend on the frequency of the larynx tone or on the speech rate, etc. With the system it is possible to use bandpass filters with comparatively broad transmission band widths. A formal description of the system by means of characteristic functions of the features allows direct design and construct of the system by means of generally available integrated circuits.