Recording and evaluating soundscapes in Singapore

Abstract

Planning urban infrastructure requires an understanding of Singapore's soundscape. A study was done to assess the various areas of Singapore based on the "Pleasantness" and "Eventfulness" of the soundscape as evaluated by the ISO models. This project was an extension of the study. This was done by recording 62 sites based on the locations of the prior study, then automatically evaluating the soundscape characteristics using a pre-trained model trained on the ARAUS dataset. Several random sets of excerpts were generated in order to find the best set of points to reflect the quadrants used in the earlier study. These excerpts were chosen based on an automated method with random seeding over a representative loss function, which was devised specially for this project.