Some investigations on noise resilient and speaker independent speech recognition

Abstract

In recent years a remarkable improvement has been seen in the performance of automatic speech recognition system Yet the performance of automatic speech recognition systems is still far below that of human speech perception capabilities Accuracy of automatic speech recognition is influenced by a number of factors say speaker variation channel distortion reverberation noise etc There has been much research on improving newlineautomatic speech recognition performance in these environments As automatic speech recognition systems are trained under laboratory conditions they generally fail to produce satisfactory performance under more adverse conditions and actual scenarios As the speech recognition and spoken language technologies are being transferred to real world applications a need for greater robustness in recognition technology is becoming increasingly apparent Bearing these in mind this research work has been focused on the newlineimplementation of noise resilient and speaker independent speech recognition system newlineFeature based techniques extensively influence the performance of speech recognition systems and also they tend to be computationally efficient and responsive to changing conditions Speech feature extraction methods have been generally modelling the human voice production system or auditory system This research work has been attempted by pursuing both these approaches and tested for the speech recognition newline newline