Dysphonia Detection Based on Voice Signals Using Naive Bayes Classifier

Voice pathology detection has gained a lot of attention in the last few decades. Furthermore, this field is considered an active topic in the healthcare area. However, most machine learning techniques are proposed to differentiate the healthy voice from the pathological voice only, where there is a lack of identification of a certain voice disease. Therefore, this work presents a method for detecting Dysphonia Disease (DD), which belongs to the pathology detection application. The proposed method uses the Naive Bayes (NB) algorithm as a classifier in order to identify the dysphonia (pathological) class from the healthy (normal) class. In addition, the Mel-Frequency Cepstral Coefficient (MFCC) is used for extracting the acoustic features. The acoustic signals used in this method were gained from the Saarbrucken Voice Database (SVD). Several evaluation measurements have been used to assess the proposed method. The experiment results indicate that the NB classifier obtained an accuracy of 81.48%, 65% sensitivity, a specificity of 91.17%, and a 76.98% G-mean. Further, the precision and F1-score are 81.25% and 72.22%, respectively.