Back     Close

Common modes of vocal tract articulation based on kinematic data
Brad Story and Anubhav Swami

Vocal tract shaping patterns based on tongue, mandible, and lip displacement data from four speakers in the University of Wisconsin x-ray microbeam database [Westbury, UW-Madison,1994] were determined with a Principal Components Analysis (PCA). Cross-distance functions representative of each of eleven vowels were first obtained from the pellet positions and the profile of the hard palate for the four speakers. The PCA was then independently performed on each speaker’s set of cross-distance functions. The results indicated that the first two orthogonal components (referred to as modes) accounted for approximately 95% of the variance in each speaker’s set of vowels. In addition, the shape of each mode was shown to be similar across the speakers suggesting that the modes represent common patterns of vocal tract shaping. In a second part of the study, a mapping was generated that linked each speaker’s mode coefficients to cross-distance functions and formant frequencies along the time course of vowel-to-vowel transitions. With this mapping it was shown that vocal tract shapes produced during time-varying speech can be accurately represented by linear combinations of mode coefficients.

Back     Close