An Environment for Word Prominence Classification in Slovenian Language

Besides phrasing, prominence is one of the most important parameters of speech prosody to model. The so called data driven approaches nowadays seem to be the appropriate solution for prosody modeling in current text to speech (TTS) systems. They allow prosodic regularities to be automatically extracted from a prosodic database of natural speech. In this paper we’ll present an evaluation of suitability for automatic word prominence classification of Slovenian language with a hierarchical approach. We classified the prominence of words into two groups, characterized by pitch movements (pitch accent) and stress (stress accent). Pitch movements have been detected from the interpolated syllable pitch contour, while syllable stress was classified from the quantity of energy in the high band of vowel spectra. We’ll also present an examination of the correlation between the hand labeled prominent words and the extracted prosody features of the mentioned two classes.