Physics-Informed Machine Learning in Sound Field Estimation: Fundamentals, State of the Art, and Challenges (video)

Sound field estimation, also referred to as sound field reconstruction, capturing, and interpolation, is a fundamental problem in acoustic signal processing that aims to reconstruct a spatial acoustic field from a discrete set of microphone measurements. Its wide range of applications includes the visualization/auralization of an acoustic field, spatial audio reproduction using a loudspeaker array or headphones, and active noise cancellation in a spatial region. Classical techniques were based solely on physical acoustics, but in recent years, there has been progress in combining them with more advanced signal processing and machine learning techniques. In particular, learning-based techniques using neural networks enable adaptation to the acoustic environment and improve accuracy when the number of available microphones is limited, and they are attracting attention as physics-informed machine learning (PIML). In this webinar, based on our recent overview article in IEEE Signal Processing Magazine, we will present the basics and recent trends in PIML-based sound field estimation techniques.