Morphological and acoustic analysis of the vocal tract using a multi-speaker volumetric MRI dataset

The shape of the vocal tract was analyzed from both morphological and acoustic perspectives for ten male speakers of Japanese. A volumetric MRI (magnetic resonance imaging) measurement was performed while each speaker uttered each of the five Japanese vowels. The cross-sectional vocal-tract area function was computed from the MRI dataset and the resulting 50 vocal-tract shapes were analyzed statistically to determine the principal deformation patterns. A perturbation of the vocaltract shape was then given for each vowel to examine the effect on the first and second formant frequencies. When the perturbation was given by changing the coefficient values of the first and second principal modes, a local region on the coefficient plane was observed where the formant change was small. In other words, this region was acoustically insensitive to the perturbation of the vocal-tract shape. When the vocal-tract shapes of the ten speakers were marked on the same plot, it was also found that marked vocal-tract shapes were located in the vicinity of the acoustically insensitive region. From these numerical investigations, it was considered how the individual differences in the vocal-tract shape can be resolved to generate phonetically relevant speech sounds.