A method of co-registering multiple magnetic resonance imaged vocal tract volumes for fricatives

In a study of fricative production, Magnetic Resonance Imaging was used to investigate three-dimensional tract geometries of fricatives produced in different vowel contexts. Scan sequences were designed to minimize the number of elicitations required, to reduce acquisition times, and to constrain the imaging of a given phone to a single session. A new method of assembling vocal tract volumes was developed to consolidate the data obtained using this approach. Sagittal, axial and oblique-coronal imaging orientations were chosen to best resolve the tract in the midsagittal plane, the pharynx, and around the fricative constriction. Each fricative token was acquired using a 36 second imaging sequence. Three corresponding point clouds were created by sampling air-tissue boundaries in each image stack, and aligned using anatomical landmarks on the face and spinal column, creating a single, multiplysampled volume. A composite tract model was developed, using the most suitable data source at each region of the tract. Vocal tract models of eight English fricatives, produced in three vowel contexts, were constructed in this manner. The models have provided important insights into the articulation of the constriction and tongue grooving, as well as differences in the production of voiced/voiceless fricatives.