ABSTRACT
This study is interested in whether there exists a predictable relationship between the mid-sagittal tongue contour and its related 3D tongue surface shape during speech. The assumption is that for any single language, a limited set of phonemically based 3D tongue shapes are used. If these shapes can be delineated and mapped to specific midsagittal displacements and cross-sectional (coronal) shapes, then predictions from midsagittal displacements to coronal shapes and then to 3D shapes can be made for specific speech sounds. The present study examined two ultrasound data sets: (1) the 3D static tongue surface reconstructions from a single subject (Stone & Lundberg, 1996), and (2) five coronal slices for two sentences spoken by a second subject (Yang & Stone, 2002). The midsagittal-to-coronal relationship for the 3D surfaces was extracted and applied to the continuous speech data. The predicted midsagittal-to-coronal relationship was well captured. This result supports the idea that a knowledge of the 3D shapes of a language, even based on a single speaker, can then be used to transform 2D midsagittal data into a 3D surface for other data sets.
