PARAFAC analysis of the three dimensional tongue shape

(Received 2 3D tongue shape during vowel production is analyzed using the three-mode PARAFAC model. 3D MRI images of five speakers (9 vowels) are analyzed. 65 virtual fleshpoints (13 segments along the rostral-caudal dimension and 5 segments along the right-left direction) are chosen based on the interpolated tongue shape images. Methods used to adjust the alignment of MRI images, to set up the fleshpoints and to measure the position of the fleshpoints are presented. PARAFAC analysis of this 3D coordinate data results in a stable two-factor solution that explains about 70% of the variance.