A new method for error correction in strings with applications to spoken word recognition

This work focuses on the problem of error-correction in strings. We propose a new method for correcting possible errors in phonemic-strings which are obtained by some preprocessing of spoken words and then applying statistical decision to resulting features. We set up a new distance measure for measuring the distance between 2 strings. The proposed method uses the new distance measure and the available linguistic information in the form of rules of the following two types; (i) rules obtained from the theory of word formation (from phonemes) in English, and (ii) rules needed to correct some errors made by statistical decision rules used earlier. In an experiment, involving 400 utterances and 40 words, an error correction rate of over 99 percent was achieved using the proposed error correction method.