Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

Voiced Diphone Synthesis Using a Parametric Model and Formant Based Mapping

Dongbing Wei, J. W. Devaney, C. C. Goodyear

Department of Electrical Engineering and Electronics University of Liverpool, Liverpool, UK

The vocal tract shapes used by a particular speaker for nine vowels, have been measured using both magnetic resonance imaging and an acoustic reflectance method. These data have been used to obtain parameter values for a simple parametric model of the vocal tract from which the areas of a 21-section acoustic tube synthesiser are found. A technique has been developed for interpolating parameter values among the nine vowel points in the f1,f2 plane. This makes it possible to obtain area functions which, when used in the synthesiser, will provide any prescribed values of the first two formants within the speaker's vowel triangle. The method effectively defines a two-dimensional subspace of the parameter space which is accessed by formant frequencies, thereby offering a method with low computational cost for articulatory copy synthesis of voiced speech.

Full Paper

Bibliographic reference.  Wei, Dongbing / Devaney, J. W. / Goodyear, C. C. (1995): "Voiced diphone synthesis using a parametric model and formant based mapping", In EUROSPEECH-1995, 1841-1844.