We propose a set of rules for the computation of prosody which are implemented in an existing generic Data-to-Speech system. The rules make crucial use of both sentence-internal and sentence-external semantic and syntactic information provided by the system. In a Textto -Speech system, this information would have to be obtained through text analysis, but in Data-to-Speech it is readily available, and its reliable and detailed character makes it possible to compute the prosodic properties of generated sentences in a sophisticated way. This in turn allows for a close control of prosodic realization, resulting in natural-sounding intonation.
|Title of host publication||Proceedings of the Workshop on Concept to Speech Generation Systems, Madrid, Spain, July 11|
|Editors||K. Alter, H. Pirker, W. Finkler|
|Publication status||Published - 1997|