Voice Pathology Detection And Classification

Abstract

Voice pathology is increasing dramatically, especially due to unhealthy social habits, being too much talkative, age factors or some kind of pathology in throat. Normally the people who speak a lot for example, teachers or announcers etc. suffer from voice pathology in their elderly age. A research-oriented simulation project is developed, which will help a general physician to identify this voice pathology. It is work of specialist ENT doctor. This will help and assist general physicians to identify pathology and to refer the patient to the specialist doctor. Automatic voice pathology detection and classification (AVPDC) systems can help general physicians detect any presence of voice disorders and the kind of voice disorders from which the individual has to suffer in the initial stages. Various disorders are taken as voice pathology detection and for classification four different voice disorders of vocal folds have been taken which are common in Pakistan i.e. Cyst, Polyp, Paralysis and Laryngitis. For implementing the work, a hybrid system is developed which is based on two models. For assisting general physicians, the model first detects the pathology i.e. normal voice or pathological voice. In the second model the system will classify the type of disease present in the patient. The main goal of this research work is to investigate commonly used features available for feature extraction of voice pathology classification and detection. Then train the system using various machine-learning techniques that can provide us with better pathology detection and classification rate. This paper focuses on developing an accurate Mel frequency Cepstral Coefficient feature extraction technique for classifying and detecting voice pathologies using a German Saarbrucken voice database. The system is trained by using four different kinds of machine learning algorithms which are Naïve Bayes, Support Vector Machine, K nearest neighbor, Random Forest and then finding out which algorithm performs well in both of these models. This project is research-oriented, so validating the results and comparing it with previous research work is analyzed. Analysis of several different diseases and techniques are also provided in these studies