Date of Award
1-29-2024
Publication Type
Thesis
Degree Name
M.A.Sc.
Department
Electrical and Computer Engineering
Keywords
Sub-node Neural Network;Voice Pathology Detection
Supervisor
Jonathan Wu
Rights
info:eu-repo/semantics/openAccess
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Abstract
The human voice carries unique individual characteristics and finds widespread application in voice pathology detection (VPD). In this thesis, human audio recordings are utilized to develop a readily accessible, rapid, and efficient methodology for the automated detection of Coronavirus 2019 (COVID-19). This endeavour represents a solution to the challenge of COVID-19 and its variants detection. The paper proposes a novel one class classification model based on the sub-node neural network (OCC-SNN). Unlike conventional approaches that incorporate multiple layers in the hidden space and employ complex techniques to determine the global optimum, the proposed network is clearer and more interpretable. In each iteration, one sub-node is added to the original structure, and the weights of the new sub-node are calculated using the Moore-Penrose (MP) inverse and the error obtained from the previous iteration. Furthermore, a detailed method of equalizing Mel-frequency cepstral coefficients (MFCCs) is proposed to extract the features from audio recordings. Furthermore, utilizing the original Coswara dataset, we manually curated optimized datasets to facilitate a more comprehensive evaluation. The proposed COVID-19 detection frame approach is able to distinguish positive sample from healthy samples, and the proposed OCC-SNN algorithm also extends to effectively image-based anomaly detection.
Recommended Citation
Li, Zeng, "Subnetwork-Based Neural Network for Voice Pathology Detection" (2024). Electronic Theses and Dissertations. 9170.
https://scholar.uwindsor.ca/etd/9170