Computer simulation of a speech-recognition system

This paper describes the use of digital speech-processing facilities to investigate sets of measurements appropriate for recognition purposes. Two parameters, namely turn arounds and zero crossings per 10ms, are chosen as particularly suitable. The reliability of these measurements is discussed in some detail, and an analysis is given of the causes of fluctuation in the speech signal. A program is described which simulates a speech-recognition system based on these parameters, and this is tested on 19 male and female speakers representing a variety of regional English accents. The results indicate real-time capability and an accuracy of the order of 96% for small command-vocabulary applications. A proposal is made for a practical hardware implementation of the scheme.