Text independent classification of normal and pathological voices using MFCCs and GMM-UBM

This paper proposes a text independent method for the classification of normal and pathological voices. If the classifier is text dependent i.e classifier is trained for a particular phoneme, then it may difficult for the patient to pronounce the particular phoneme. To overcome this difficulty, a text independent classification method is proposed, which uses Mel-Frequency Cepstral Coefficients (MFCCs) and Gaussian Mixture Model-Universal Background Model (GMM-UBM). The GMM-UBM model is trained with phonemes /a/, /e/ ,/u/ of normal and pathological voices. Hence the classifier is efficient to detect voices of different phonemes and classifies them into normal and pathological with a maximum accuracy of 85.63% . It has been noticed that, accuracy of classification can be improved by increasing the number of MFCCs, i.e the classification accuracy is 72.45% for 12 MFCCs , where as 85.63% for 24 MFCCs.