Parameter estimation for spectral matching in articulatory synthesis

A major difficulty in generating natural sounding speech using articulatory synthesis is the lack of sufficient data on the parameters of the synthesiser and the motions of the articulators. The author describes a method for estimating the parameters of the nasal section of a vocal tract model, using input speech spectra estimated using a conventional technique. The use of a neural net in estimating the coupling into the nasal tract is also described. >