4th International Conference on Spoken Language Processing

Philadelphia, PA, USA
October 3-6, 1996

Speaker Individualities of Vocal Tract Shapes of Japanese Vowels Measured by Magnetic Resonance Images

Chang-Sheng Yang, Hideki Kasuya

Faculty of Engineering, Utsunomiya University, Ishii-machi, Utsunomiya, Japan

Three dimensional vocal tract shapes of three males and three females were measured from the magnetic resonance images that were taken during sustained phonation of the five Japanese vowels. Dimensional differences in the vocal tract length of the subjects were quantitatively measured by dividing the entire vocal tract into the oral, the pharyngeal and the laryngeal sections. To investigate main factors that contribute to the differences of the formant patters, uniform and non-uniform normalization methods were applied to the vocal tract shapes. The formant frequencies were also computed from the normalized area functions and compared. The results show that the physiological dimensions of the vocal tract continuously distributed from females to males. The normalization experiments suggest that the non-uniform scaling of the vocal tract was not significant and back/ front cavity volume in addition to the vocal tract length is an important factor to describe speaker individualities.

Full Paper

Bibliographic reference.  Yang, Chang-Sheng / Kasuya, Hideki (1996): "Speaker individualities of vocal tract shapes of Japanese vowels measured by magnetic resonance images", In ICSLP-1996, 949-952.