A Fast Approximate Acoustic Match for Large Vocabulary Speech Recognition

Abstract

In a large vocabulary speech recognition system, a great deal of computer time is spent performing detailed acoustic matches of words in the vocabulary. In order to run in real time on a modest amount of hardware, it is important that these detailed matches be performed only on words which have a reasonable possibility of being correct. We describe a scheme for rapidly obtaining an approximate acoustic match of all of the words in the vocabulary in such a way as to ensure that the correct word is, with high probability, one of a small number of words examined in detail. We give experimental results showing the effectiveness of the fast match that we describe for a number of talkers. In particular, we give the average rank of the correct word, the average number of words suggested, and the number of times that the correct word is not among the words suggested.