Publication

Abstract : Voice disorders affect communication and quality of life, requiring effective diagnostic tools for early detection. The proposed work employs machine learning techniques to analyze various voice pathologies using the Saarbrücken Voice Database (SVD). Diseases such as Hyperfunctional Dysphonia, Contact Pachydermia, Reinke's Edema, Spasmodic Dysphonia, Chordectomy, Dysodia, and Vocal Cord Polyp are examined. Different models, including K-Nearest Neighbors (KNN), XG-Boost, Support Vector Machine (SVM), and Logistic Regression, are evaluated for their accuracy in detecting voice disorders. The impact of data augmentation using the Synthetic Minority Oversampling Technique (SMOTE) is analyzed to improve model robustness. Results show that XGBoost achieved the highest accuracy, with 87% for male voices and 89% for female voices. For individual diseases, the best accuracy with SMOTE was 98% for Chordectomy, using SVM and XGBoost models. SMOTE consistently improved accuracy across most conditions, especially for underrepresented voice disorders.