Poster: Resource-Efficient Environmental Sound Classification Using Hyperdimensional Computing

Run Wang, Shirley Bian, Xiaofan Yu, Quanling Zhao, Le Zhang, Tajana Rosing

Abstract: On-device environmental sound classification (ESC) in rural areas faces one major challenge of resource efficiency. Traditional methods rely on resource-intensive machine learning models, making them impractical for small edge devices like microcontrollers (MCUs). This poster presents SoundHD, a novel ESC solution using Hyperdimensional Computing (HDC), a brain-inspired and lightweight computing paradigm. We further optimize the memory footprint for deployment on MCUs. Our initial results show that SoundHD can be deployed and executed effectively on memory-constrained MCUs.