Automatic Detection and Annotation of Sperm Whale Codas

TL;DR

This work presents the first automatic detector and annotator for sperm whale codas, addressing the lack of scalable tools for automatic detection, annotation, and source separation of codas in passive acoustic data. It introduces a graph-based clustering framework that leverages the similarity of multi-pulse coda clicks, with structural and temporal likelihood analyses to distinguish codas from echolocation and to assign clicks to the same coda across overlapping sources. The approach enables large-scale coda characterization, revealing synchronized exchanges between whale pairs, discovering two new coda types, and enabling quantitative analyses of coda type distributions, inter-coda timing, and information transfer in codas. The method’s demonstrated robustness at low SNR and across near-field and far-field recordings supports real-time deployment for crewed and autonomous platforms, advancing both cetacean communication research and conservation management.

Abstract

A key technology in sperm whale (Physeter macrocephalus) monitoring is the identification of sperm whale communication signals, known as codas. In this paper we present the first automatic coda detector and annotator. The main innovation in our detector is graph-based clustering, which utilizes the expected similarity between the clicks that make up the coda. Results show detection and accurate annotation at low signal-to-noise ratios, separation between codas and echolocation clicks, and discrimination between codas from simultaneously emitting whales. Using this automatic annotator, insights into the characterization of sperm whale communication are presented. The results include new types of coda signals, analyzes of the distribution of coda types among different whales and for different years, and evidence for synchronization between communicating whales in terms of coda type and coda transmission time. These results indicate a high degree of complexity in the communication system of this cetacean species. To ensure traceability, we share the implementation code of our coda detector.

Figures (13)

• Figure 1: Illustration of a scenario with a focal and a non-focal whale emitting 5-click codas. An example of a super-resolution spectrogram of individual clicks from a real tag recording of the codas of the focal and non-focal whale are shown in the top right and bottom right panels, respectively. A clear distortion of the structure of the clicks can be seen in the coda received from a distance. However, the similarity between the clicks is retained.
• Figure 2: ROC curve, evaluated for near- and far-field data. The solid lines refer to the unconstrained detection mode, while the lines for the constrained mode are dashed. False alarm rate evaluated for noise data including a) sperm whale echolocation clicks, b) ambient noise without echolocation clicks (only far-field data).
• Figure 3: Cumulative distribution function (CDF) of the ratio of the number of clicks detected in the coda in the near- and far-field datasets using constrained detection mode.
• Figure 4: An example of 13 seconds from a near-field recording showing a dyadic coda exchange between a focal and a non-focal whale. The Inter Coda Interval of the focal and non-focal whale are indicated by black solid and dashed arrows, respectively. The Inter Coda Break is indicated by red solid arrows.
• Figure 5: Analysis of the distribution of coda types for all dyadic coda exchanges found. The left panel shows the distribution matrix of all pairs of coda types, with each bin representing a particular coda type from the focal whale (signal) and the subsequent coda type of the non-focal whale (response). The right panel is zoomed in on the most active coda types. The bins' color represents the probability of occurrence on a logarithmic scale.