Chapter 8 - Classifying Recurrent Dynamics on Emotional Speech Signals

Abstract

The current study was an attempt to understand the dynamics of emotional speech signals. The nonlinear dynamics of two speech signals having two different emotions were investigated using recurrence analysis. First, two types of speech signals were collected when the subject spoke eight sentences in the Bengali language in angry and normal emotion. The recurrence plot and the phase space plot were constructed for each signal and recurrence-based features, namely %isometry and %consecutive isometry were extracted. A marked distinction was visually observed between the normal and the angry emotion-based signals from the recurrence and the phase space plots. The variation was further confirmed from the values of the recurrence parameters. The normal emotion exhibited higher %isometry and %consecutive isometry as compared to the angry emotion. The robustness of the recurrence-based method for classifying the emotional states was established from the consistency of the results provided by the recurrence parameters in both noise-free and noisy environments.