Voice Pathology Detection Using Convolutional Neural Network and Data Augmentation

Abstract

Detecting and identifying voice disorders, which frequently result in diminished vocal quality, is essential for effective treatment and improved patient outcomes. This paper focuses on developing a Deep Learning-based approach to identify voice pathologies. The proposed approach combines the use of Convolutional Neural Network (CNN) with data augmentation techniques. Mel Frequency Cepstral Coefficients (MFCCs) are utilized as input features to the CNN model to perform binary classification of voices (normal or pathological). Experiments were conducted using two types of voice recordings from the Saarbrücken Voice Database (SVD), including vowels and sentences. The evaluation results demonstrate the superiority of our approach in in terms of accuracy for pathological voice detection compared to SVM appraoch and KNN techniques.