A transputer based speech recognition system

The authors discuss the application of transputers to the development of a powerful and flexible prototyping environment for investigating applications of speech recognition. Two practical systems are described to illustrate the way in which transputer arrays can be configured to concentrate computing power where it is most needed. The approach allows rapid algorithm development and testing in the Occam high-level language. Strong compiler type checking and run-time error checking engender a high level of confidence in the system performance. Real-time response is available by matching the number of processors to the complexity of the task, thus, for example, allowing the perceived performance of the recognizer/dialogue combination to be assessed for specific applications. The availability of flexible processing power limited only by cost allows many ideas to be investigated within a practical time scale