Computational Perception Group

home | members | publications | directions | research | presentations | teaching | vacancies | personal

A statistical, formant-pattern model for segregating vowel type and vocal-tract length in developmental formant data

R. E. Turner and T. C. Walters and J. J. M. Monaghan and R. D. D. Patterson

Published in: Journal of the Acoustical Society of America


This paper investigates the theoretical basis for estimating vocal-tract length (VTL) from the formant frequencies of vowel sounds. Magnetic resonance imaging (MRI) of the vocal tract shows that the growth of the components of the vocal tract is highly non-uniform, so the formant ratios of vowels might be expected to vary considerably with speaker height. Moreover, analysis of the variability in formant frequencies suggests that there might well be another component of variability beyond vowel type and VTL. A statistical model was developed to describe the relationships between vowel type, VTL, formant frequency, and vocal cavity size. When the parameters of the model are learned from formant data, the model shows that VTL is the major source of variability after vowel type, and that the contribution due to developmental changes in oral-pharyngeal ratio is small relative to measurement noise. This suggests that speakers adjust the shape of the vocal tract as they grow to maintain specific formant-frequency ratios for individual vowels, and as a result, it should be a straightforward matter to estimate VTL from vowel sounds, given vowel type.

bibtex, pdf


 
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback