Estimations of Vocal Area Functions during Vowel Production from MRI Data and Speech Signal

Abstract

We can utter sounds by changing a shape of our vocal tract. A vocal tract characteristic has been analyzed by AR model for speech signals and a vocal tract area function can be estimated by PARCOR coefficients computed from the AR model. On the other hand, a method has been developed for three-dimensional reconstruction of the vocal tract shape from magnetic resonance imaging (MRI). In this paper, the vocal tract area function during vowel production will be estimated based on the following two methods and they are compared.

　(1) It is estimated as PARCOR coefficients from AR model for speech signal.

　(2) It is calculated as cross section area of vocal tract extracted from MRI data acquired while a subject is uttering the long vowel.