Title: Spectral descriptors for the assessment of vocal fold nodules and feature optimisation using MRMR algorithm

Authors: Jennifer C. Saldanha; Malini Suvarna; Dayakshini Sathish; Cynthia Santhmayor; Rohan Pinto

Addresses: Department of Electronics and Communication, St. Joseph Engineering College, Vamanjoor, Mangaluru, India; Affiliated to: VTU Belagavi, India ' Department of Electronics and Communication, Tontadarya College of Engineering, Gadag, India; Affiliated to: VTU Belagavi, India ' Department of Electronics and Communication, St. Joseph Engineering College, Vamanjoor, Mangaluru, India; Affiliated to: VTU Belagavi, India ' Department of Speech, Father Muller College of Speech and Hearing, Kankanady, Mangalore, India ' Department of Electronics and Communication, St. Joseph Engineering Collge, Mangalore, India; Affiliated to: VTU Belagavi, India

Abstract: Objective assessment of voice in vocal nodules from the spectral descriptive features is discussed in this study. Further, MATLAB feature engineering is explored to automate the process of feature engineering. The performance of a set of optimisable classifiers such as decision tree, bagged trees, Naive Bayes, linear and quadratic support vector machines was evaluated on feature engineered dataset. The decision tree outperformed all other classifers with an accuracy of 84.2% for engineered features. Spectral centroid obtained highest ranking using maximum relevance and minimum redundancy feature ranking method and found to be most appropriate for classification. Harmonic ratio, harmonic to noise ratio, shimmer variants, and spectral centroid features obtained a significant amount of correlation with the perceived degree of hoarseness. Among these features, spectral centroid is found to be strongly negatively correlated, hence can be effectively used as a quantitative indicator to measure the level of severity of the pathologic voice.

Keywords: maximum relevance minimum redundancy; MRMR; harmonic to noise ratio; HNR; noise to harmonic ratio; NHR; cepstral peak prominence; CPP; correlation feature selection; CFS; support vector machines; SVMs; artificial neural network; ANN.

DOI: 10.1504/IJISTA.2025.148888

International Journal of Intelligent Systems Technologies and Applications, 2025 Vol.23 No.3, pp.337 - 362

Accepted: 10 Sep 2024
Published online: 30 Sep 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article