Prosodic prominence detection in Italian continuous speech using probabilistic graphical models

Prosodic prominence, a speech phenomenon by which some linguistic units are perceived as standing out from their environment, plays a very important role in human communication. In this paper we present a study on automatic prominence identification using Probabilistic Graphical Models, a family of Machine Learning Systems able to properly handle sequences of events. We tested the most promising members of such models on utterances selected from a manually annotated Italian speech corpus, obtaining very good recognition results crucially converging with the prominence detection responses provided by a pool of native speakers.