We have attempted to apply modern linguistic theory, specifically non-linear phonology, to the task of text-to-speech conversion. We believe that approaches to text-to-speech conversion based on such linguistic theory offer greater possibility of improved naturalness in synthetic speech. This improvement is made possible by the ability of non-linear techniques to represent accurately the complex structures underlying natural speech, and also to supply an element of cognitive modelling to the task. This paper presents an outline description of the non-linear framework around which the proposed text-to-speech conversion system is to be constructed and the feature driven formant synthesis technique underlying the implementation.
Keywords: Non-linear, non-segmental, speech synthesis.