Confidence measure driven scalable two-pass recognition strategy for large list grammars

In this article we will discuss recognition performance on large list grammars, a class of tasks often encountered in telephony applications. In these tasks, the user makes a selection from a large list of choices (e.g. stock quotes, yellow pages, etc). Though the redundancy of the complete utterance is often high enough to achieve high recognition accuracy, large search space presents a challenge for the recognizer, in particular, when realtime, low latency performance is required. We propose a confidence measure driven two-pass search strategy, exploiting the high mutual information between grammar states to improve pruning efficiency while minimizing the need for memory.