Hybrid Approach for Analyzing Acute Spots of Clinical Speech Data Using Fuzzy Inference System

Abstract

Acoustical measures of vocal functions are used in the assessments of voice disorders and monitoring the subject's improvement with speech therapy. In this chapter, a hybrid approach is proposed to identify the acute spots in pathological speech signals. These spots represents where the speech disorder occurs. The speech training for that specific portion of speech in particular could be given for enhancing the speeches. Dimensionality reduction is done using Principal Component Analysis (PCA) on Mel Frequency Cepstrum Coefficients (MFCC) extracted. By statistical method it is proved that overall 91.60% of the words were classified correctly. The features were trained using Support Vector Machines (SVM) for categorizing normally and abnormally pronounced words. The peaks found by Fast Fourier Transform (FFT) in abnormal words is made use of in the Fuzzy Inference System (FIS) for finding the acute spots in which the aberration has occurred in the word. This hybrid approach was found to have around 98% accuracy.