Abstract
Understanding cetacean whistles is crucial for assessing their social interactions, behaviours, and responses to anthropic activities. Identifying the various types of whistles present in acoustic recordings is often challenging, but necessary for this purpose. To facilitate this process, we have developed a semi-automated deep learning approach called the "Draw Your Own Contours" (DYOC) method. This is available as an open-source software along with its associated dataset. It utilises YOLOv8m and ResNet18 to identify whistle contours. DYOC was applied to 808 minutes of audio recordings of wild, free-ranging short-beaked common dolphins from the Bay of Biscay, France. It enabled the annotation of 8,730 whistle contours, six times faster than manual annotation. These recordings were associated with observations of dolphin behaviour, the presence of fishing nets, and the activation of the DOLPHINFREE acoustic beacon. Analyses revealed that these variables affected the signal-to-noise ratio, the number of inflections, and the frequency and/or duration of recorded whistles. This study provides the first characterisation of whistle features for a population of short-beaked common dolphins in the Bay of Biscay. The annotation of whistle contours using the DYOC method helped reveal the complex acoustic behaviour of dolphins in response to external variables.