Evaluation of prosodic juncture strength using functional data analysis

Abstract

Prosodic structure has large effects on the temporal realization of speech via the shaping of articulatory events. It is important for speech scientists to be able to systematically quantify these prosodic effects on articulation in a way that is capable both of differentiating between the degree of prosodic lengthening associated with varying linguistic contexts and that is generalizable across speakers. The current paper presents a novel method to automatically quantify boundary strength from articulatory speech data based on functional data analysis (FDA). In particular, a new derived variable—the Deformation Index—is proposed, which is the area under FDA time-deformation functions. First using synthetic speech produced with the TaDA task dynamics computational model, the Deformation Index is shown to be able to capture a priori known differences in boundary strengths instantiated in the π-gesture framework. Additionally, this method accurately distinguishes between types of boundaries in non-synthetic speech produced by four speakers.