Model-based formant-to-area mapping with labial cross-section constraints

The object of the presentation is model-based formant-to-area mapping. The objective is the examination of the morphological similarity between observed and mapped area functions that may involve constraints on the labial cross-section. The vocal tract model is a concatenation of truncated cones the cross-sections of which are inferred by means of an inverse mapper that is based on a locally linear relation between observed formant frequency increments and calculated cross-section increments. The corpus comprises eight speakers who have sustained 77 vowels. Results show that constraining the labial cross-section improves the similarity between recorded and calculated area functions. The increase is less than 10 %, however. Also, fixing the labial cross-sections for French rounded back vowels, the area functions of which are difficult to infer from acoustic data, does not qualitatively improve the similarity between observed and inferred area functions.