Novel Method of Nonlinear Symbolic Dynamics for Semantic Analysis of Auditory Scenes
Pauline Mouawad, Shlomo Dubnov · 2017
Discovering semantic information from complexsignals is a task concerned with connecting humans'perceptions and/or intentions with the signal's content. In the case of audio textures from environmental sounds produced bya cheering crowd, laughter, crackling fire, car crash orexplosion, complex meanings of the events are inferred andappraised in a listener's mind, which triggers an affective response that is relevant for well-being and survival. In this paper we contribute to the ongoing research on affective semantics from sound by proposing a novel learning framework of affective auditory scene analysis using a recently developed method of non-linear dynamic signal analysis. Using an adaptive symbolization process that finds the best audio structure representation in terms of the symbolized sequence recurrence properties, we show thatmeasures of periodicity and complexity derived from our model are relevant for the characterization of affect in auditory scenes, and that they will perform better than state-of-the-art methods relying on low-level acoustic features.