Improving word recognition accuracy by means of syntax

This paper discusses the effect of an optimal type of syntax analysis on isolated word recognition accuracy for a specific speech recognition task. The results were obtained by a computer simulation of the system. The results show that for an acoustic word error rate of 10%, the syntax analysis reduces this to 0.2% which gives a sentence error rate of 1%. These figures are for 1000 randomly generated sentences with an average length of 10.3 words per sentence and a 127 word vocabulary. We believe that these results are indicative of the performance which will be obtained from real speech recognition systems based on the method of syntax analysis described herein.