Abstract: Confidence measures enable us to assess the output of a speech recognition system. The confidence measure provides us with an estimate of the probability that a word in the recognizer output is either correct or incorrect. A confidence measure is a quantitative estimate of a word’s correctness. There are several approaches that incorporates one or two earlier methods to quantify the performance of confidence measures. It can be applied to improve recognition by incorporating extra information into the recognition process or weighting hypothesized words. It can also be used for predicting recognition accuracy and detection of recognizer failure. This study presents an approach that incorporates confidence measures based on both acoustic model and language model for measuring the word-based recognition reliability.