Measuring Inter-Pulse Intervals in Sperm Whale Clicks: Development of an Automatic Method and its Potential for Identifying Eastern Caribbean Social Units

Abstract

Sperm whale clicks have a unique multi-pulsed structure, where the inter-pulse interval (IPI) is related to body length. This feature makes passive acoustic monitoring especially informative for sperm whales. I investigated methods for analysing IPIs automatically from raw audio recordings. First, I developed an algorithm for compiling reliable IPI measures. I then used this algorithm in an attempt to identify social units of sperm whales in the Eastern Caribbean.

In practice, IPIs are difficult to measure, because the multi-pulsed structure is clear only when clicks are recorded along the whale’s longitudinal axis. To achieve automatic IPI compilation, I trained a support vector machine (SVM) to recognize probable on-axis sperm whale clicks. The compilation algorithm uses the SVM to isolate “Good” clicks, whose IPIs are then validated based on precision and repetition. This method was found to be successful in producing IPI distributions with precise peaks that likely corresponded to individual whales. These peaks could also be resolved using Gaussian mixture models, thereby providing automatic estimates how many whales were present and how large they were. However, the routine rejected a considerable number of clicks (> 99% on average). Nevertheless, since sperm whales click regularly, audio recordings lasting at least 10 minutes are likely to yield adequate IPI distributions for the vocalizing whales.

The second objective of my thesis was to automatically identify permanent sperm whale social units (with about 3-12 members each) based on IPIs. I established IPI profiles for 5 units commonly encountered off Dominica, and used these to infer unit presence from empirical IPI distributions. IPI profiles showed some potential for recognizing individual units, but the routine successfully identified all units present in only about 30% of cases. Groupings of units and the presence of unknown individuals were particularly problematic.

The automatic IPI compilation algorithm developed here should make it possible to assess length distributions of sperm whales from passive acoustic surveys, even when multiple whales are clicking simultaneously. IPIs also have potential for identifying social units, which would be particularly useful for monitoring the declining Eastern Caribbean population. However, the unit detection routine implemented here needs to be improved.