Fuzzy voice segment classifier for voice pathology classification

Speech is one of the common modes of communication and it is a process of transferring information from one entity to another. In recent years there has been much research on unvoiced/voiced classification and voice pathology classification. In this research work a simple fuzzy classifier has been designed to segment the voiced and unvoiced portions of a speech signal. A simple feature extraction algorithm is proposed to extract the Tri Mean relative average perturbation (Tri Mean-RAP) features from the segmented voice portion of the signal. Further, using PCA transformation the significant Tri Mean-RAP features are extracted and a simple neural network model is developed. In the proposed fuzzy classifier, the energy per frame and change in energy level between the adjacent frames are fuzzified and rules are formulated to segment the voiced portion. The Tri Mean-RAP features are then extracted from the segmented voice portion. The proposed methods are validated through simulation.