Recognition of English vowels using top-down method

Abstract

Many recognizers use bottom-up methods for recognizing each phoneme or feature, and use the cues and the context to find the most appropriate words or sentences. But humans recognize words not just through bottom-up processing, but also top-down. In many cases of listening, one can usually predict what will come based on the preceding context, or one can determine what has been pronounced by listening to the following sounds. Therefore, if some cues to a word are given, it would be possible to refine the recognition by using the top-down method. This thesis deals with the improvement of the performance of recognition by using the top-down method. And most of the work will be concentrated on the problem of vowel recognition, when the adjacent consonants are known.