Acoustic Evaluation of Simplifying Hypotheses Used in Articulatory Synthesis

International audienceArticulatory synthesis allows the link between the temporal evolution of the Vocal Tract (VT) shape and the acoustic cues of speech sounds to be investigated unlike other approaches of speech synthesis. A number of simplifying assumptions have to be made to enable the speech signal to be generated in a reasonable time. They mainly consist of approximating the propagation of the sound in the vocal tract as a plane wave, approximating the 3D VT shape from the mid-sagittal shape, and simplifying the vocal tract topology by removing small cavities. This work is dedicated to the evaluation of these assumptions. For this purpose 3D MRI VT shapes for sustained vowels and fricatives were acquired together with the speech sound recorded via an optical microphone. Vocal tract resonances were evaluated from the 3D acoustic simulation computed with the K-wave Matlab package from the complete 3D VT shape and compared to those of real speech, those provided by simplified 3D VT shapes (near epiglottis), and those provided by either the electric simulation from the mid-sagittal shape, or the 2D K-wave package from a mid-sagittal VT shape. We will exploit these results to improve the simplified acoustic simulations used in articulatory synthesis