Sound source localization in a natural soundscape with autonomous recorder units based on a new time-difference-of-arrival algorithm

• Sound source localization is useful for animal behavior tracking and ecology assessment. • Localizing a sound source outdoor is challenging due to noise and high calling rate. • A deterministic algorithm combines rough separation and time of arrival computation. • The procedure was test on a 190 min recording of a complex freshwater soundscape. • The time–space components including amphibian, avian and human sources was determined. Acoustic sound source localization is an emerging approach in animal behavior and ecology to either monitor individuals or to describe the spatial and temporal structure of natural soundscapes. The localization of sound sources mainly rely on the deployment in the field of microphone arrays and on the subsequent pairwise comparison of signals to infer position through triangulation. However, most of the current methods face the challenge of separating or not the sound sources before estimating their spatial position. On the one hand, source localization with upstream separation is particularly challenging in noisy outdoor environments. On the other hand, no-separation sound source localization can be possible only when calling rates are low. Here, using four outdoor low-cost recorders, we propose a new method which is an intermediate between the separation and no-separation strategies. Our deterministic algorithm combines a rough call separation, the computation and self-consistency check of time difference of arrivals, the separation or aggregation of sound sources, the localization of the sources in Cartesian or polar coordinates, and the final identification by an expert. We tested this method on a 190 min soundscape recording achieved in a temperate freshwater environment. The automated analysis revealed the space and time pattern of an amphibian, avian and human complex soundscape. This procedure opens the possibility to deploy similar passive acoustic programs to monitor either predetermined individuals or the composition of dynamic soundscapes.