Classification of Healthy and Pathological voices using MFCC and ANN

The automatic system for classification of healthy and pathological voices has received a significant attention in the research of early detection and diagnosis of voice disorders. In this work, we propose a method to classify the healthy and pathological voices. To implement this system, we use audio recordings of normal and pathological voices. We extract Mel Frequency Cepstral Coefficients (MFCC) from the voice signals and use a visualization technique to explore the capability of these features in discriminating healthy and pathological voices. In this study, we use Artificial Neural Network (ANN) to classify the extracted features. Here, we present the results of experiments with varying number of neurons in the hidden layer and also with various frame sizes. The best obtained accuracy is 99.96%.