Speech is the most natural and quick mode of transforming and sharing information. To automate the process of speech production and perception, many researches are carried out for more than five decades. For an Automatic speech Recognition (ASR) of a large or unlimited vocabulary, a recognition unit smaller than word size is necessary. In this paper a new approach for segmenting the input utterance into individual characters is presented. The accuracy of boundary detection of baseline Maximum Likelihood (ML) Algorithm and the proposed algorithm is also compared and discussed.

No abstract