Building Health Monitoring Using Computational Auditory Scene Analysis

Abstract

This paper presents a method to identify sound sources for structural monitoring, known as building health monitoring. This method allows to evaluate deterioration and damage of buildings by analyzing environmental sounds. The proposed method determines the location and features of sounds generated within a building, with its main characteristics being: (1) planar direction and height estimation; (2) visualization of sound features according to loudness, continuity, and pitch. The capabilities of the proposed building health monitoring method are verified using environmental sound data acquired at a building in Gunkanjima, which is a world heritage site from Japan.