The shift-ACF: Detecting multiply repeated signal components

We propose a novel method for detecting multiply repeated signal components within a source signal. Whereas standard methods such as classical autocorrelation are usually tailored to detecting signal components that are repeated once, our approach, the so called shift-autocorrelation, can be used to improve detection performance by explicitly assuming multiply repeated components. In this paper we introduce the shift-autocorrelation formally and give theoretical evidence of its performance compared to classical autocorrelation. By defining tempograms based on shift-autocorrelation, we apply the method to the detection of underwater mammal sounds, showing its superiority to classical approaches in a practical evaluation.