An exploratory study of emotional speech production using functional data analysis techniques

Abstract. Speech articulations associated with emotion expression were investigated using electromagnetic articulography (EMA) data and vocal tract data acquired using a fast magnetic resonance imaging (MRI) technique. The data are explored using functional data analysis (FDA) techniques for articulatory timing and vocal tract shape analyses. It is observed that the EMA trajectories of tongue tip movements associated with the production of a target word segment are quite similar across emotions examined in this study, suggesting that articulatory maneuvers for linguistic realization are largely maintained during emotion encoding. Results of the functional principal component analysis of the vocal tract shapes also support this observation. Mainly, the articulatory movement range and velocity (i.e., the manner of articulation) are modulated for emotional expression. However, another interesting articulatory behavior observed in this study is the horizontal shift of tongue tip positioning when emotion changes. Such a strategy by a speaker may facilitate the task of emotional contrast as long as such a variation of the place of articulation does not obscure linguistic contrast. [Supported by NIH, ONR-MURI] 1.