Noise-Robust Algorithm for Determination of Prosodic Characteristics of Speech Signals in Automated Assessment System of Human Psycho-Emotional State

An automated noise-robust algorithm for determining prosodic characteristics of speech is proposed. The nature of the algorithm consists in decomposing a speech signal into informative noise and signal frequency components by means of improved complete ensemble empirical mode decomposition with adaptive noise, and selection of the component containing the pitch with the subsequent determination of prosodic characteristics. A study was conducted using a verified base of pure and noisy speech signals recorded in 220 males and females, aged 18 to 79 years, with signs of psycho-emotional disorders. In accordance with the results of the study, the algorithm provides good noise robustness of various intensities (signal-to-noise ratio is from 0 dB to 30 dB), and can be successfully tested in actual conditions of an “aggressive” noise environment for automated assessment of a human psycho-emotional state.