A comparison study of emotional speech articulations using the principal component analysis method

Abstract

In this study, we will investigate differences in the tongue movements across emotions by utilizing the principal component analysis (PCA), which enables to detect major and minor variations in the entire tongue surface movements under different speaking conditions. For the purpose of the study, we analyze an acted emotional speech production corpus collected from one actor and two actresses using an electromagnetic articulography (EMA). Discrete emotion types considered in this study are anger, sadness, and happiness as well as neutral one as reference. Specifically, we will investigate the number of principal components that are needed to capture emotional variations, and the differences in tongue shaping across different emotion types. Outcome of the study would provide supplementary information to the previous PCA-based production studies which have mainly focused on normal, or neutral, speech articulation. Major principal components that are found in the study also can be utilized as a basis by which effective but compact articulatory correlates (i.e., component scores) can be derived for a further investigation of emotional speech production such as an interaction between articulatory kinematics and prosodic modulations of pitch and loudness patterns, which is another important motivation of the study.