Study of soundscape in Singapore and its correlation to urbanization policies, with a focus on noise modelling and noise perception

Abstract

Noise pollution has been increasingly focused upon due to their severe impact on health. However, little widespread study has been done to analyse noise in urban context, specifically in Singapore. In this paper, we take an investigation to understand what are the common noises heard in Singapore, and analyse how noise is perceived. Specifically, this study focuses on three main aspects: i) Crowdsource information based on primary noise data collected; ii) Train a sound classification model that can classify audio files collected in Singapore; iii) Generate findings on how human perceive different kind of noises. An online survey was conducted to understand how humans label and perceive different audio file. The collected information was used to determine the audio file labels and the corresponding files were used to train two kinds of deep learning model – the Multi-Layer Perceptron (MLP) and Convolutional Neural Network (CNN). The MLP model were explored and trained to achieve the maximum accuracy policy. Both the MLP model and CNN model have an agreeable accuracy at 77% and 72% respectively, and can be used to predict audio files from Singapore. Secondly, analysis of noise perception showed that loudness alone may not be a factor in people perceiving audio files negatively. Analysis of audio frequency shows that variation in pitch correlates more with negative perception instead of a specific pitch range. This study has initialized a deep-dive into connecting noise to urban policies. With the model trained and the data collected, further studies can be conducted to link its findings with other socioeconomic factors to correlate with urban policies.