Atom decomposition based stress detection and automatic phrasing of speech

Abstract

The Weighted Correlation based Atom Decomposition (WCAD) is a recently proposed physiological intonation model that decomposes the pitch contour into elementary components - atoms. Since these atoms are said to correspond to laryngeal muscle activation, in theory they could be used to infer higher linguistic meaning from the pitch contour. One such application relevant for cognitive infocommunication is the automatic detection of stress that we evaluate on the Hungarian language, for which this task is equivalent to phonological phrasing. The results show high correlation between stress and the presence of an atom peak in a syllable, as well as between stress and/or pauses and atom valleys in the previous syllable. We have run further experiments to evaluate the atom decomposition based stress detection against hand-made stress annotations and a Hidden Markov Model based stress detection system. The analysis shows comparable results with the two methods.