Robust detection of multiple bioacoustic events with repetitive structures

In this paper we address the task of robustly detecting multiple bioacoustic events with repetitive structures in outdoor monitoring recordings. For this, we propose to use the shift-autocorrelation (shift-ACF) that was previously successfully applied to F0 estimation in speech processing and has subsequently led to a robust technique for speech activity detection. As a first contribution, we illustrate the potentials of various shift-ACF-based time-frequency representations adapted to repeated signal components in the context of bioacoustic pattern detection. Secondly, we investigate a method for automatically detecting multiple repeated events and present an application to a concrete bioacoustic monitoring scenario. As a third contribution, we provide a systematic evaluation of the shift-ACF-based feature extraction in representing multiple overlapping repeated events.