The paper describes the design of a text-to-speech oriented model of Italian intonation, based on analyses of natural speech and implemented as the pitch assignment module of an actual synthesis system for the Italian language. The model was not intended to be an exhaustive description of natural intonation, its aim being the generation of plausible melodies, appropriate to neutral reading style. The experimental investigations, concerning fundamental frequency and its relations with stress and syntax, were carried out on a subset of a speech database explicitely designed for prosodic analyses. Sentence pitch profiles were 'stylized', following the guidelines of the perceptual approach developed at the Institute of Perception Research of Eindhoven. The standardized pitch movements and their recurring configurations were organized in a set of rules associating f0 variations to text structures. The sketched model was implemented in the Olivetti text-to-speech system, replacing a previous pitch algorithm, and was augmented and refined by direct tuning. The naturalness of synthetic speech resulted considerably improved.
Bibliographic reference. Quazza, S. (1991): "Modelling Italian intonation in a text-to-speech system", In EUROSPEECH-1991, 1161-1164.