Modelling coarticulation

An outline is presented of the hypothesis underlying a speaker-centroid-based representation of vowels. The representation suggests parallels across different sources of variation, including speaker category, and segmental and suprasegmental contexts. In all cases, the author first locates the origin of the vowel space at the speaker centroid. Individual vowels are represented by a vector of orientation θ and magnitude r. Across the sources of variation outlined above, the angle of orientation seems to remain fairly stable. Context- or speaker-independent variants in vowel formants rest on modification by a single factor across the entire system of r. The speaker-centroid based representation is thus potentially very powerful.