Analyzing and Correction of Errors to Improve the Speech Recognition Accuracy for Telugu Language

Dramatic advances have been made to improve the accuracy in the area of Speech Recognition. Many differences in user articulation, acoustic environment, transmission channel, microphone, word collection likelihood, dialectal variation, speech rate-all caused reduced performance of the speech recognition system. Mis recognition may occur due to substitution of single word for another word. Confusable words which leads to less recognition accuracy for the large vocabulary speech recognition system. Hence to improve the speech recognition accuracy, error recovery procedures are essential to use in the speech recognition system. Modification of the dictionary will be done based on the confusion words which are obtained from the decoder of the speech recognition system. Mismatch occurs when new acoustic signal is given to the decoder which is different from the training system. Testing speech models are not consistent with the acoustic and language model of the speech which is used in the training phase. In such cases, statistical method will be applied to the output of the decoder. Error Correcting procedures are essential and imposed by analyzing the errors that occurs in the decoder of the speech recognition system.