Study of Voiced Speech Using Empirical Mode Decomposition to Detect Stressful Emotions in Human-Robot Interaction

A method for studying voiced speech by means of improved complete ensemble empirical mode decomposition with adaptive noise to detect stressful emotions in human-robot interaction is presented. The informative parameters of voiced speech fully reflect the violation in functioning of organs of human speech apparatus due to emotional excitation. The key feature of the research is to decompose a signal into empirical modes, to detect modes containing periodic information on the excitation source of the vocal tract, and to construct a composite signal reflecting information on glottal activity. The research results of voiced speech, being computation of the fundamental frequency of 100 multi-harmonic signals, are presented. The pitch frequency modulation of multi-harmonic signals in the frequency range of 0-2.5 Hz/ms in 0.5 Hz/ms increments simulated the irregularity in vibration of the vocal folds (30-40 % of the nominal value) arising from the emotional excitation of a person. The results obtained allow us to conclude that the presented method for the study of voiced speech, using the improved decomposition, can be successfully tested to detect stressful emotions in human interaction with robotic devices.