A Rule-Based Italian Text-to-Speech System

An Italian text-to-speech system based on the INFOVOX architecture will be described- With respect to earlier versions of the system, the complete set of Italian vowels and diphthongs was taken into consideration. A set of specific rules for the assignment of word stress, entirely based on statistical considerations, as well as a complete set of grapheme-to-phoneme rules were implemented. Appropriate parameter definitions for the realisation of Italian phonemes were formulated and, on the basis of preliminary results on the study of Italian prosodic structure, appropriate phonetic changes were introduced. Finally, the results of perceptual experiments which were carried out in order to evaluate the system will be described.