Empowering our Critters: Running Energy Efficient Deep Learning Models for On-Edge Bioacoustic Monitoring

We propose a multi-phase plan to study the feasibility and benefits of deploying energy-efficient deep learning models on edge devices using passive acoustic monitoring (PAM) for climate change mitigation and adaptation. Current approaches involve collecting audio recordings for long periods and then automatically annotating them with bioacoustic models at a central location; we make the case for running these bioacoustic models on-edge to significantly speed up the operational efficiency of PAM. Successful deployments would open up several possibilities, including early warning systems for climate disasters, frequent monitoring of a region's resistance to extreme climate events, and real-time monitoring of the effects of human activities on ecosystems.