Microprocessor based automatic speech recognition system for flight applications

Abstract

Advances in the semi-conductor industry have made possible significant improvements in speech recognition hardware especially in the areas of size, power consumption, and cost. A microprocessor based discrete word recognizer to be used in various flight applications is described and preliminary performance results given. The system utilizes a time warping algorithm based on spectral change to normalize the duration of spoken inputs and dynamic updating of the stored patterns to reduce the effects of slow changes in the speakers voice. The entire preprocessor including shaping circuits, filter bank, multiplexer, A/D and all digital processing including CPU, random access memory, program memory, and input/output circuitry is contained on a single 612 × 912 in. printed circuit board. Word prototype pattern storage and recognition algorithm logic is implemented externally, either through another microprocessor based system or a standard minicomputer. In this way, one slow and relatively inexpensive computer may handle several simultaneous speech inputs in real time and at a fraction of the cost of currently available equipment.