Akademik Veri Yönetim Sistemi
3rd Global Conference on Engineering Research, 13 - 16 Eylül 2023, ss.335-346
Voice is a kind of energy that is shaped according to pressure changes, which is
effective in the production of structures such as tongue, teeth, palate and larynx, as well as
vocal cords. Voice, which has a variable dynamic in its nature, is also a unique identification
tool for each person. The vocal folds produce the voice as vibrating by air passes through them
during exhalation of air from the lungs. Some different deformation in these vocal tract
components, especially vocal cords, which make up the voice in pathological cases are
reflected in the voice, and the structure of the voice changes. Therefore, voice analyzes by
using acoustics and traditional analysis techniques are frequently used in the assessment of
vocal pathologies. In this study, healthy and three vocal cord pathologies such as cyst, polyp
and sulcus were investigated by using wavelet transform and extracting features some of them
come from nonlinear approaches which give useful especially findings in voice analysis.
Feature selection algorithms of the Relief, one way Analysis of Variance (ANOVA) and
Minimum Redundancy Maximum Relevance (MRMR) were performed to reduce the number
of features by determining the importance of them. A four-class classifying procedure was
realized by using two classifiers of Cosine K-Nearest Neighbor (KNN) and Neural Network to
detection of voice disorders. Experimental results demonstrated that the multi-class
classification test accuracies of 88.6 and 88.0 were achieved for Neural Network and KNN
classifiers respectively and MRMR selection algorithm. The results provided promising
classification accuracies in four-class classification to determination of voice diseases of vocal
cord.