A Method for Automated Segmentation of Speech Signals to Determine Temporal Patterns of Naturally Expressed Psycho-Emotional States

A method for automated segmentation to determine temporal patterns of speech signals reflecting naturally expressed psycho-emotional states has been developed. The nature of the method is the segmentation of speech into informative areas using adaptive decomposition and energy analysis of intrinsic mode functions, and determining patterns of temporal intervals of voiced, unvoiced, and pause sections. A research using the formed base of speech signals of 100 subjects, experiencing natural positive and negative emotions, was conducted. The research results were evaluated in comparison with the known methods of segmentation, followed by the determination of temporal patterns. In accordance with the determination results of psycho-emotional states, it was concluded that the developed automated method more accurately determines the boundaries of informative areas, due to the advantages of the energy analysis of empirical modes obtained by the method of improved complete ensemble empirical mode decomposition with adaptive noise.