Machine learning / deep learning approach to soundscape evaluations

Abstract

Masking is the addition of sounds to soundscapes or noise-polluted areas. These additional sounds are known as “maskers”. Soundscape augmentation is a method that involves the addition of “maskers” to a soundscape. It is a noise mitigation method that aims to improve the overall soundscape perception or quality. Many studies have used such techniques to improve the perception of a soundscape. However, the studies conducted have some limitations. The choice of maskers used in those studies are often limited to a single type of masker and are inflexible to real-time soundscapes. The method for selecting maskers also tends to be dependent on experts. This project will be using a machine learning/deep learning approach to select maskers from the given masker database for a soundscape, which can instantaneously and independently predict a suitable masker for that soundscape to create an overall pleasant soundscape.