Detection of clicks based on group delay

In this paper we present a novel approach for the automatic detection of clicks from recordings of beaked whales based on the phase characteristics of minimum phase signals and especially using the group delay function. Group delay is estimated through the and first derivative of the Fourier Transform of a signal. A major advantage of the proposed approach is its robustness against additive noise while it doesn't require the definition of ad-hoc or adaptive thresholds for the detection of clicks. This method works on raw recordings which are usually quite noisy as well as on click enhanced recordings (after band-pass filtering or using operators like the Teager-Kaiser energy operator). Moreover, a click is just detected by searching the positive zero crossings over time of the slope of the phase spectrum. To evaluate the effectiveness of the proposed approach in detecting clicks, a one-minute recording has been manually marked providing a test set of about 320 clicks. Results show that the proposed approach was able to detect 71.37% of the hand labelled clicks within an accuracy of 3 ms.