First results for a speaker-independent refutation based speech recogniser designed to operate in open system environments

The natural environments in which practical speech recognition devices are required to operate are open systems in the thermodynamic sense (Bertalanfy 1971). This means that they can exhibit an increase in order. Hence their states will be non-denumerable. For the purpose of evaluating automatic speech recognisers it is necessary therefore to distinguish between a finite set of known classes of utterance such as is given by a finite recognition system vocabulary and an infinite set of comparable entities whose particulars are unknown. It has bee argued by others and is here assumed to be true, that all Pattern Matching approaches to Automatic Speech Recognition are logically equivalent. Pattern Matching is based on verification principles: the greater the evidence in favour of a given hypothesis, the higher the probability of that hypothesis being correct