Multi-speaker validation of coarticulation models of syllabic nuclei

The acoustical dynamics within English CVC-syllables is modeled by a set of phonetically-motivated parametric functions of the formant-frequency trajectories of vocalic nuclei. The notion of linear superposition of contextual effects leads to an approximate decomposition of vowel-formant transitions into a constant vowel target, an initial CV- and a final VC-transition. The latter transitions are further modeled asL + kf(n)andL' + k'g(n), wheref(n)andg(n)are per-consonant common-contour shape functionskandk'are scaling constants, andLandL'are interpreted as consonant loci. These models were tested on a small database produced by three native Australian male speakers, who recorded five repetitions of Australian English /CVd/-syllables, where C=/h,b,d,g,p,t,k/ and V=/i,I,ε,æ,þ,λ,3,a, / in /CVd The techniques for constructing these models from formant data are presented and fits to multi-speaker test data analysed.