Evaluating Syntactic Constraints to Speech Recognition in a Fighter Aircraft Environment

摘要

A flexible software system has been developed to test the effects of adding syntactic knowledge to an isolated speech phoneme-based word recognizer. Words from seventy-word fighter plane vocabulary, spoken by five pilots at four different levels of background noise, are automatically concatenated into commands randomly chosen from a set of over seven trillion. These commands are then recognized using an existing word recognizer together with grammars of differing specificity. Results are compiled automatically. The system is flexible in that system components such as the command generator, parser, grammar, or word recognizer can be interchanged with very little software modification. Preliminary testing demonstrated that, although the modified word recognizer exhibited very poor performance, the use of more specific grammars enhanced recognition accuracy, sometimes drastically. (Author)