Stress detection: A sonic approach

Abstract

We propose a transformation of spoken signals that retains the prosodic characters of speech, and is robust. Most of the existing signal processing for this purpose is based on the assumption that the signal be continuous in time. Our transformation changes the recorded signal to a discontinuous one, highlighting silences and their time structure, dropping any semantic content not related to rhythm and intonation. The (numerical) transformation is "sonic" in that it is to be judged by trained ears, in the sense of native speakers of a language. We set forward a conjecture for stresses occurring after implosive consonants in Italian and Brazilian. The perceptual synchronization that characterizes it, is probably a more general phenomenon regarding also stresses in other contexts. © Springer-Verlag Italia, Milano 2009.