Phoneme-Level Parameterization of Speech Using an Articulatory Model

An articulatory domain representation of the speech signal is desirable for the following reasons: the parameters are likely to interpolate over long intervals even through consonants; a high degree of acoustic detail can be reproduced from simple constraints governed by physical laws; the glottal source and vocal tract interactions, and voice-unvoiced transitions, have the potential for matching those in natural speech; and for text-to-speech application, it may be easier to formulate rules in the articulatory domain. However, articulatory data are hard to obtain and this has prevented the widespread use of articulatory parameters from the speech signal and also demonstrate that good quality synthesis is possible with parameters specified once per phoneme.