Harnessing the Power of Deep Learning for Urban Sound Sensing and Noise Mitigation

In the digital age, the integration of sensing, processing, and sound emission into IoT devices has made their economical deployment in urban environment possible. These intelligent sound sensors, like the Audio Intelligence Monitoring at the Edge (AIME) devices deployed in Singapore, operate 24/7 and adapt to varying environmental conditions. As digital ears complementing the digital eyes of CCTV cameras, these devices provide public agencies with a wealth of aural data, enabling the development of comprehensive and effective sound mitigation policies. In this presentation, we will examine the critical requirements for intelligent sound sensing and explore how deep learning techniques can be utilized to extract meaningful information, such as noise type, dominant noise source direction, sound pressure and frequency of occurrence of the environmental noise. Additionally, we will introduce new deep-learning-based active noise control and mitigation approaches, including reducing noise from entering residential buildings and generating acoustic perfumes to mask annoyance in urban environments; and how these deep learning models can be deployed in an edge-cloud architecture. Our aim is to demonstrate how deep learning models can advance the field of acoustic sensing and noise mitigation and highlight current challenges and trends for future progress.