Third International Conference on Spoken Language Processing (ICSLP 94)
We have developed an accurate method to measure vocal tract (VT) shape and length from magnetic resonance (MR) images acquired during sustained phonation of Japanese vowels. The 3-dimensional (3D) VT shape was reconstructed by using coronal MR images for the oral cavity and axial MR images for the pharyngeal and glottal regions. A mid-sagittal image was used as a reference for the reconstruction. MR images of dental impressions of a subject were also incorporated into the reconstruction. All the MR images were directly transferred to a workstation where an intercative program was used to measure the VT shapes. Three Japanese subjects, a child, a female and a male, participated in the experiment. Formant frequencies were computed from the 3D VT shapes which were measured for the three subjects assuming one-dimensional sound wave propagation through the vocal tract. The first three formant frequencies were compared with the ones measured directly from real voice. Except for a few cases, differences between the two measurements were all less than the difference limen (DL) of the formant frequencies.
Bibliographic reference. Yang, Chang-Sheng / Kasuya, Hideki (1994): "Accurate measurement of vocal tract shapes from magnetic resonance images of child, female and male subjects", In ICSLP-1994, 623-626.