Machine learning / deep learning approach to soundscape evaluations

• Main Author: Wong, Arthur Jun Xiang
• Other Authors: Gan Woon Seng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering
• Online Access: https://hdl.handle.net/10356/177228
• Institution: Nanyang Technological University

Soundscape augmentation is a paradigm shift in noise mitigation by placing greater emphasis on the human perception of the acoustic environment. This is performed by introducing additional sounds to mask the background noise for better acoustic comfort. Evaluation of such data is crucial for the understanding of the diversity of responses for different demographics. This study aims to improve the IoT subsystem for adaptive soundscape augmentation, which is done in another study called ‘Deployment of an IoT System for Adaptive In-Situ Soundscape Augmentation’ [1], by classifying demographic-sensitive perceptions of sound through the incorporation of live age and gender detection. Deep learning models, based on MiVOLO [2], are employed to accommodate the variations in height, angles, and distances. The robustness of the model is then validated through a series of controlled experiments, employing a diverse dataset to ensure inclusivity and accuracy.