Articulatory speech synthesis using a parametric model and a polynomial mapping technique

A model of the vocal tract has been developed that allows the complete area function to be expressed efficiently in terms of nine quasi-articulatory parameters. Sets of parameters obtained from magnetic resonance imaging have been used to seed diphone codebooks, which are then filled using a parameter inching technique. A least squares technique is used to obtain a polynomial approximation to the mapping from formant frequencies to synthesiser parameters. It has been found that a cubic mapping is sufficient to obtain a reasonable approximation