The reliability of speech recognition systems can be enhanced considerably through the use of pre-defined wordsets and phraseology syntaxes. Historically, syntax structures have been defined manually using hand-drawn state diagrams, which in turn are converted manually to ASCII files. When many words, nodes or connections between nodes are involved, the processes of defining, debugging, and modifying a syntax can be quite tedious. A tool has been developed which automatically generates a graphical state diagram from an ASCII syntax definition file. It can also check for various hazards in the structure, check a list of phrases for compliance with the syntax, count words, and write and check against word dictionaries. This automates much of the clerical tedium of dealing with syntax structures and phrase lists. A further enhancement, which allows graphical editing of the state diagram, and subsequent automatic generation of the descriptive ASCII file is in the design phase. An additional real-time nodal flow analyzer is also included in the tool package.

With the development and addition of post-processed phraseology checking and word scoring/word thresholding utilities, apparent voice system accuracy and user acceptability may be significantly increased. This paper discusses the use of these speech recognition tools and utilities in the NTSC Under Ice Navigation Trainer Test-bed.