Keyword Feedback for Improving Speech Recognition in Command and Control Information-Acquisition Tasks

摘要

Our objective in the Speech Technology Group of the User Interface Technology Branch at the Naval Ocean Systems Center (NOSC) in San Diego is to apply speech technology to naval command and control systems. By employing data bases and expert systems to aid in decision making, naval command and control systems are becoming more complex and sophisticated. The design of man-machine interfaces to these systems is critical to user acceptance and functionality. We believe that the use of speech recognition in many of these systems is not only possible, but vital to shipboard operations. This report addresses a number of important issues involving the implementation of continuous speech recognition for information-acquisition tasks. The following questions are addressed: 1) For a restricted information acquisition task such as database query, what can be considered natural speech, 2) How can the syntax be constrained to improve accuracy but at the same time still allow natural speech, 3) How effectively can feedback be used to shape or influence the user's speech. Can the user adapt to a limited vocabulary and restricted syntax, 4) For an appropriate task, can a continuous speech recognizer configured as a word spotter be used to partially remove syntax restrictions. An information acquisition experiment is described. Results are presented and user interaction and reaction are discussed. (AW)