Skip to main content
2023 ESA Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Contributed Talk

Session: : Remote Sensing And Image Analysis 4

COS 311-4 - A fully automated method for acoustic identification and localization of terrestrial wildlife using autonomous recording units

Thursday, August 10, 2023
4:15 PM – 4:30 PM PDT
Location: E144
Abstract: The acoustic monitoring of wildlife at large spatial and temporal scales has been enabled by the increasing availability of affordable hardware and developments in machine learning techniques. However, large-scale acoustic deployments are typically not able to collect the fine-grain spatial data necessary to measure microhabitat use, map territories of individuals or use spatial methods like distance sampling to estimate abundance. Small-scale arrays of microphones have been successfully used to localize sound sources - but the technical challenges of precise time-synchronization of audio, sound detection and classification, and localization of sound sources have limited the scale at which acoustic localization arrays have been used. Here, we present an integrated, open source hardware and software framework for synchronized recording, classification and localization of terrestrial wildlife using acoustic autonomous recording units. We use battery powered AudioMoth acoustic recorders that have been modified to accommodate GPS receivers, and developed software that uses the resulting GPS time-stamped audio to precisely time-synchronize recordings taken across many autonomous recorders. We trained Convolutional Neural Networks (CNNs) for detection and classification of sounds of interest, used generalized cross-correlation phase transform (GCC-PHAT) to find time delays of arrival (TDOAs) between filtered audio from receivers, and used multilateration methods to localize sound sources from these TDOAs. The entire software pipeline of training an audio classifier, synchronizing GPS time-stamped audio recordings, filtering and cross-correlation of audio and localization of sound sources will be integrated within our open source Python package OpenSoundscape. We validated the performance of our localization framework using two separate field tests: a small array tested with loudspeakers playing bird songs from precisely known locations, and a true field test deployment of an array of over 60 recorders in the acoustically dense setting of an Eastern Deciduous Forest in May. In the true field test we validate the results of our localization framework against the results of simultaneous spot mapping surveys conducted at the same site. Our framework successfully localizes singing birds in dense soundscapes and enables the use of acoustics to collect fine-grain spatial data on acoustically detectable wildlife at the landscape scale.