Analysis of Dravidian words uttered by deaf speakers using clustering techniques: deaf speech enhancement

Abstract

A novel method for grouping the fundamental speech features of normal and deaf speakers using a new Modified Self Organising Map (M-SOM) clustering algorithm, and deaf speech enhancement using distance measure is presented. The M-SOM algorithm will automatically categorise the normal and deaf speaker speech features into two different clusters. Its performance is analysed in comparison with SOM and NMTF algorithms. The result obtained reveals that the M-SOM algorithm which is developed by combining the adaptive features of SOM and the Ward clustering methods provides lowest intra and highest inter-clustering and appears to be the best method for deaf speech signals. After clustering estimation of distance metric, termed as correction measure facilitates enhancement of the unclear deaf speech so that it could be understandable to normal speakers. Hence it can be inferred that the suggested method effectively enhances the deaf speech signal for better understanding by a normal listener.