Acoustic source identification in noisy environments
Date of Award
2024
Document Type
Thesis - ECU Access Only
Publisher
Edith Cowan University
Degree Name
Master of Engineering Science
School
School of Engineering
First Supervisor
Iftekhar Ahmad
Second Supervisor
Daryoush Habibi
Abstract
Acoustic Source Identification (ASI) has many industrial and environmental applications, and the associated techniques and systems are continually improving in accuracy and efficiency. The detection of acoustic sources in noisy environments has been a topic of interest for researchers and scientists in the past few years. Identifying acoustic sources has been useful in many industrial and military applications, including acoustic ranging, acoustic surveillance and navigation, robot-nature interaction, and hearing aids that visualise sounds. However, it is difficult to identify a sound source when multiple similar sources are present in a particular scenario. In addition, it is highly challenging to identify multiple acoustic sources present in a noisy environment. This research aims to investigate and develop an Artificial Intelligence (AI)-based ASI system for the recognition of acoustic sources in noisy environments. The two-fold research is conducted to address the research challenge using supervised and unsupervised AI methods and signal processing.
In this thesis, a comprehensive survey of AI-based ASI in various applications is presented in which the strengths and weaknesses of AI-based ASI processes and associated methods proposed by researchers in the literature are analysed and compared. This concept is applied to sounds originating from terrestrial vehicle sources, with pronounced sound feature correlations affected by wind noise. Acoustic data from vehicles is collected, pre-processed, and MFCC feature vectors are extracted for training supervised Machine Learning (ML) algorithms, such as Support Vector Machines (SVM), K-Nearest Neighbours (KNN), and Random Forests (RF). The trained ML models are tested under challenging scenarios to evaluate the extent of training and recognition performance of models. As a result of testing and evaluation, SVM achieved the highest 89.28% recognition accuracy in the first case, while RF outperformed in the second case achieving 84.44%.
Moreover, the work is extended to address the acoustic source separation in underwater environment. An underwater noise model is presented based on Gaussian distribution and Raleigh distribution to investigate the performance of the proposed approach. In order to effectively separate multiple underwater sounds and reconstruct them, Blind Source Separation (BSS) and denoising of underwater acoustic signals are performed. The unsupervised ML algorithm Non-Negative matrix factorisation (NMF) separates acoustic signals from noisy signal mixture followed by Minimum Mean Square (MMSE) to reduce the influence of water area noise in the reconstructed signals. Results demonstrate the effectiveness of the approach by reducing the interference of noise in separated acoustic signals. Compared to stateof-the-art BSS techniques such as iFastICA and conventional NMF, the proposed method improved separation performance by minimising the Mean Square Error (MSE) between the clean signal and reconstructed signal by up to 47.5% and 36.0%, respectively.
DOI
10.25958/v4my-p795
Access Note
Access to this thesis is embargoed until 23rd August 2025.
Recommended Citation
Zaheer, R. (2024). Acoustic source identification in noisy environments. Edith Cowan University. https://doi.org/10.25958/v4my-p795