Microphonemes as Fundamental Segments of Speech Wave Primary Segmentation - Automatic Searching for Microphonemes

Abstract

This paper concerns the stage of acoustic analysis in speech recognition of Speech Understanding System SUSY; subsystem AKORD, Acoustic analysis in this system is based on dynamic spectral analysis of the speech wave. It corresponds with the function of the hearing organ expressed by an analog nodel of human ear. Acoustic analysis in the AKORD system uses an FFT algorithm but is not strictly based on accepting constant intervals of 10-20 ms; analysis does not cause inaccuracy at the level of parametrization because of a considerable dispersion of parameters of particular segments. A conclusion was made in the AKORD system that a properly conducted process of segmentation will help in solving some problems of speech wave analysis, particularly recognition and time compression. This paper attempts to choose the primary segment (microphonem) in the most optimum way and to Indicate the algorithm of automatic segmentation based on microphonemes as the dynamic segments.