Word-based confidence measures as a guide for stack search in speech recognition

The Maximum a posteriori hypothesis is treated as the decoded truth in speech recognition. However, since the word recognition accuracy is not 100%, it is desirable to have an independent con dence measure on how good the maximum a posteriori hypothesis is relative to the spoken truth for some applications. E orts are in progress [1, 2, 3] to develop such con dence measures with the intent of applying it to assesment of con dence of whole utterances [4], rescoring of N-best lists, etc. In this paper, we explore the use of word-based con dence measures to adaptively modify the hypothesis score during search in continuous speech recognition: speci cally, based on the con dence of the current sequence of hypothesized words during search, the weight of its prediction is changed as a function of the con dence. Experimental results are described for ATIS and SwitchBoard tasks. About 8% relative reduction in word error is obtained for ATIS.