The design and operation of the mechanical speech recognizer at University College London

The design of a practical speech recognizer, incorporating many of the principles discussed in a companion paper, is described. The principal parts of the recognizer are as follows:– acoustic spectrum analyser; spectral pattern matcher; store of information about linguistic statistics; phoneme sequence memory; the computer; output circuits (a) the storage of computer decisions, (b) the control of the automatic typewriter. The speech material used with the completed recognizer is described, together with the results obtained. Future developments in speech typewriter research and possible uses of a successful recognizer are indicated.