What about the text? Modeling global expressiveness in speech synthesis

We present in this paper some modeling issues for expressive speech generation from a text necessarily determined by some extrinsic characteristics for natural output (such as the textual genre, the reader´s profile and the discursive situation). We then argue for a five-level model for main speech expressiveness, which is in the very basis of symbolic coding, formal transformation and physical generation of some prosodic patterns one meets in today´s oral French. The model and the associated algorithmic approach are developed with respect to three different phonological levels (syllable, syntagm and text) whose scope aims at covering typical micro and macro expressive phenomena