Abstract
The accuracy of Input speech signal is very essential as the speech interface is now desired for several domains. This paper attempts to propose a standard procedure for speech recognition. It also propose an algorithm to outline the cause for errors in recognition based on several related works and suggest an error recognition and repair procedures . The research work suggests that it is practically possible to predict the misrecognized utterances with a high degree of accuracy from 1. an utterance's sound file, 2. the language model being employed, 3. and recognizer outputs such as confidence, In addition, there is empirical data upon which to base successful repair strategies in relation to these misrecognitions.
