Artificial intelligence monitoring at the edge for smart nation deployment

Abstract

Using embedded devices to run artificial intelligence (AI) applications locally has been increasingly popular due to their fast real-time inferencing speeds and low power consumption. One example of an AI application that can generate valuable insights for multiple industries is sound detection and monitoring. In the education sector, noise is often considered a distraction that can contribute to school stress and impede students’ learning abilities. Thus, performing sound monitoring can assist school authorities in formulating noise control strategies to create a more conducive learning environment for students. Past research works have proposed using embedded devices to measure sound levels at various parts of schools to determine whether the school environment is suitable for learning. However, to effectively tackle the sources of disruptive noise in schools, knowing the sound levels alone is inadequate and identifying the type of sound would be a better solution. In this project, we deployed an AI application on an embedded device to record surrounding sounds, perform real-time sound classification, and display the classification results on a cloud platform for monitoring. The quantized sound classification model executed on our embedded device uses the MobileNetV2 model architecture and achieved an accuracy of over 80%. Overall, the noise monitoring embedded device for schools will serve as a helpful guide for the education sector to improve the state of their school environments.