Word recognition in continuous speech with background noise based on posterior probability measure

To overcome the issues of speech recognition, most of the existing research work considers pre-treatment techniques (noise reduction, end point detection) combined with the training models and pattern recognition techniques. To find the target word in a predefined speech sentence, various similarity measure techniques have been developed. However, the existence of background noise in speech signal degrades the performance of the system in terms of target word identification. In this paper, a novel technique for Word Recognition in Continuous Speech with Background Noise (WRBN) is proposed which differentiates the background noise from speech signal. The proposed technique uses Posterior Probability Measure (PPM), used in image processing for target localisation. Unlike PPM, background speech components which participate actively in mismatching or misidentification, have not been considered in the literature.