Reduction of confusion pairs on different rates of speech in Telugu language

Abstract

Research in speech recognition area has made considerable progress in achieving the task with tremendous growth of technology. Speech rate is one of the important factors which affect the speech recognition accuracy. In the present work, training is performed on different speech rates (Normal, Slow and Fast) and testing also done on different rates of speech. Error rate will increase when the major variations occur in building the models in training and testing. Due to these variations, more number of confusion pairs occurs and this affects the speech recognition accuracy. The aim of the present work is on the reduction of confusion pairs to recognize more number of words of different rates of speech. Reconstruction of original components will be taken place in the resultant of the decoder of the speech recognition system. Considerable amount of improvement is observed in the recognition accuracy from the experimental results. © 2013 IEEE.