French Prominence, a probabilistic framework

Identification of prosodic phenomena is of first importance<br />in prosodic analysis and modeling. We introduce in this paper<br />a new method for automatic prominence labelling. The<br />proposed method is based on well known machine learning<br />technics in a three step procedure : feature extraction, feature<br />selection for finding the more relevant prominence acoustic<br />correlates and gaussian mixture model for predicting prominence