Apparatus and method for speech recognition

摘要

A template-matching speech recognition system has a plurality of reference templates each having a plurality of addressable positions, each position storing a value corresponding to the probability of the occurrence of a particular binary value in a corresponding position in an utterance template for speech represented by the referenced template. A system bias template having a corresponding plurality of positions for storing values therein that are representative of the probability of the particular binary value occurring in a corresponding position of any utterance template. A match between an utterance template and a reference template is determined when the sum of the stored probabilities from the reference template positions, corresponding to the positions in the utterance template containing the particular binary value, exceed a predetermined threshold.