Syntax Problems with Speech Recognition in Simulator Training Systems,
Abstract
The arriving age of artificial intelligence and expert systems in military operations creates the need for a better means of entering and retrieving data. Speech recognition has for some time been a candidate for such data entry but due to its embryonic status requiring the isolated word mode, it has been deemed too cumbersome and error prone to be useful. The arrival now of truly continuous speech systems, with high accuracy, rapid data entry, and interactive query capabilities should revive serious consideration of speech recognition for this purpose. One of these new systems, recently announced by Verbex, a division of Exxon Enterprises, has all the capabilities to accomplish this task. This system, called the Model 3000, was specifically designed for continuous speech. It does not have any particular limitation on the entry of long sentences except as to how long you can speak without catching your breath; our record is 100 digits in about 30 seconds. In an application on the inspection of printed circuit boards, single utterances such as substitute resistor trim between transistor zero three lead one five and u three four pad five six are common. This paper will endeavor to show some of the techniques used to create the grammars (syntax) to accomplish difficult speech recognition tasks, based on the capability of the Model 3000.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 31, 1984
- Accession Number
- ADP003031
Entities
People
- T. Cutler