Articulatory models of Arabic vowels computed from magnetic resonance images

This paper presents the preliminary results of a long term project aimed at providing a database of articulatory models for the Arabic language. The articulation of 31 Arabic sounds was captured, and the three main Arabic vowels were studied using medical images of the vocal tract from four subjects. Magnetic Resonance Imaging (MRI) was used in this study to capture the shape and positions of speech articulators. The low temporal resolution of MRI limited this and most speech studies to sustained sounds. The images were processed, segmented, and the vocal tract models were computed. The resulting area functions are used for the calculation of formant frequencies. The results show expected natural intra-subject variabilities, but are generally closely matched and in line with the theory of vowel production. The computed models were used to generate the transfer functions and synthesize artificial Arabic vowel sounds. The quality of the synthesized sounds was confirmed to be intelligible