Application of Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise in Speech Signal Processing

Abstract

The improved complete ensemble empirical mode decomposition with adaptive noise is proposed for solving speech signal processing tasks. Systematization of speech signal processing methods is shown structurally, demonstrating the variety of solution approaches. A method for segmenting speech into informative sections is presented to determine naturally expressed psycho-emotional states of a person. The key idea of the method is decomposition of speech, energy analysis of empirical modes, and determination of temporal informative parameters of voiced, unvoiced and pause sections. The method was investigated using the generated database of speech signals per 100 subjects, experiencing natural positive and negative emotions. The research results were evaluated in comparison with the known methods of segmentation. In accordance with the results of determination of psycho-emotional states, it was concluded that the proposed method more precisely defines the boundaries of informative sections, due to the advantages of the energy analysis of empirical modes obtained by the improved complete ensemble empirical mode decomposition with adaptive noise.