A First Step Towards Understanding Sperm Whale Communication and Behavior

Abstract

We present two tools for annotating and preparing animal audio and movement data for downstream modeling tasks. We present a modification of VisionTransformer which detects and classifies animal communication into an annotated format. We give a visualization tool to create estimates for animal movement patterns and a platform for experts to assign behavioral labels to time periods. As a use case example, we show experiments training a factored Hidden Markov Model to discover patterns in the animal communication, both across the entire dataset of communication, and also split across different dominant behavior patterns. In this work each section uses sperm whales as the species of focus, however the work is easily adaptable and applicable to any species that we can collect audio or movement data for.