Automatic Speech Recognition: a Problem for Machine Intelligence

Speech recognition by machine has not yet been achieved because no suitable specification of the recognition process has been formulated for the machine. The author outlines the disturbances and constraints found in speech, and goes on to a description of the structure implied by the constraints. This description is a prerequisite of speech recognition for two reasons, first to describe general speech structure in terms which allow knowledge of it to be built into the machine as an aid to the recognition process, secondly to allow a good enough description of the input signal for it to lead to a minimum set of recognition possibilities which includes likely alternatives. The outline is drawn of a hypothetical machine to recognise speech, comprising a basic recogniser working on short segments of acoustic waveform only, on to which may be added further structures to use knowledge of speaker characteristics, speech statistics, syntax rules, and semantics, in order to improve the recognition performance. Some detailed examples of possible structures are given. Finally there is a brief description of work in progress at Standard Telecommunication Laboratories towards implementing a basic recogniser of the type suggested.