Formant-contour models of diphthongs : a study in acoustic phonetics and computer modelling of speech

This thesis advances understanding of the dual vocalic nature of diphthongs by addressing the fol'lowing question in the acoustic phonetic domain: does the vocalic duality arise simply from a static combination of two vowels as the traditional biphonemic representation of diphthongs tends to suggest, and how is the duality manifest in the time-varying formant-patterns of diphthongs? This fundamental question regarding diphthongs has hitherto not been addressed nor answered satisfactorily from either methodological or conceptual viewpoints. In particular, the sparse-sampling approach used consistently in previous studies to describe the contours of the first three formants (Fi, F2 and F3) of diphthongs attests to a long tradition which does not properly acknowledge the time-varying characteristics of the movement from onset to offset vowels. Moreover, basic understanding of the diphthong has progressed little beyond the conventional matching of two vowels, one to the onset and the other to the offset points of the Fiand Frcontours of the diphthong. In contrast, the approach adopted and advocated in this thesis attends to the detailed time-varying behaviour of not only the Fi and F2 , but also the F3 of diphthongs which are shown to play, jointly with the F2 , an important role in identifying two distinct parts of these ?peech sounds. To this end, a new method of formant estimation is employed, which secures temporal continuity along the Fi-, Fr and F3-contours of vocalic sounds. Mathematical models are then offered which (1) capture the essential contour behaviour of the first three formants of diphthongs; and (2) yield parameters which convey information about the dual vocalic nature of these speech sounds, not only in static but also in dynamic terms.