A Data-Driven Phoneme-Specific Analysis of Articulatory Importance

Speech can be seen as a result of temporally overlapping gestures of articulators comprising lips, jaw, tongue and velum. The spectrum of acoustic speech signals is influenced by these articulatory gestures. In the articulatory space, few articulators exhibit minimal variance at their target position for a specific phoneme, which are called critical articulators. The positions of these critical articulators govern the acoustic characteristics of that particular phoneme articulation. This results in a categorical decision on the articulators as critical or non-critical articulators ( Jackson and Singampalli (2009)). Instead of assigning a binary decision on articulators being critical or noncritical, in this work we attempt to assign an articulatory importance value between 0 to 1, in a data-driven manner. We denote it as articulatory importance function (AIF). Experimental analyses are performed on 38 subjects’ acoustic-articulatory data and AIF values are reported. Findings from this analysis could contribute to the understanding of inter speaker variability in speech production and variability across different languages.