Analyzing Voice Quality with Multi-Dimensional Voice Program for Disease Determination

Abstract

Multi-Dimensional Voice Program (MDVP) parameters are very popular among physicians/clinicians to detect vocal pathologies and analyze various diseases of the vocal cords. In this paper, voice pathologies are automatically detected using the parameters of the MDVP. However, MDVP is commercial software, so this work is trying to build MDVP using Python to extract MDVP parameters useful for various experiments, automatic detection of sound pathologies, and automatic classification of voice recognition. This study evaluates MDVP parameters and applies the XGBoost model as a classification method to analyze and classify diseases. This work considers three sample data, polyps, nodules, and Reinke edema, popular in clinical vocal cords diseases, from Saarbruecken Voice Database (SVD) for data testing and training. Test results demonstrate the excellent ability of MDVP’s parameter extraction to identify healthy voices and obtain accurate classification results to discriminate between healthy voices and pathological voices. The best overall accuracy is 98% using the XGBoost classifier.