ABSTRACT
Magnetic resonance imaging (MRI) has been successfully used in speech research, with the development of high B-value units and ultra-fast acquisition sequences, to obtain morphological and dynamic data of the vocal tract. This knowledge is useful for the understanding of speech mechanisms and can, in particular be employed for speech synthesis. The aim of our study was a combination of the morphological and dynamical characterizations of speech, resulting in the production of 3D articulatory models of some relevant Portuguese sounds and syllables. A set of static images in stacks and dynamic sequences were collected during the experiments from subjects that were instructed to vocalize during the acquisition, exploring the capabilities of the equipment. Analysis of the image stacks allowed the extraction of the vocal tract contours and a subsequent non-conventional 3D reconstruction by means of combination of orthogonal stacks allowed visualization and partial measurement of the sampled vocal tract shape.
