Open-source real-time passive acoustic tracking of beaked whales for low-power devices

Real-time tracking of multiple underwater sound sources remains a challenge in passive acoustics, particularly for low-power autonomous systems. Many existing tracking methods rely on post-processing or computationally intensive techniques, limiting their applicability to embedded devices. To overcome these limitations, we present a time-difference-of-arrival-based tracking algorithm that tracks multiple sources in the direction-of-arrival (DOA) domain. A key architectural feature of the method is the decoupling of continuous per-frame DOA estimation from source enumeration, enabling these processes to execute on parallel threads and ensuring that source enumeration does not degrade real-time throughput. We implement the algorithm in an open-source c++ software package, freewilli: Free software for Whale Identification and Localization with Low-Power Implementation. We demonstrate its effectiveness on passive acoustic recordings of diving beaked whale groups from a small volumetric four-channel hydrophone array deployment and show through a systematic parameter sweep that real-time processing latency remains stable across a wide range of detection rates and clustering configurations. The system operates at low power consumption (<1.3 W) on a Raspberry Pi Zero 2 W, enabling long-duration autonomous deployments.