Method of speech recognition

Claim 1

In a speech recognition system for successively judging the similarities between test patterns of speech and a plurality of stored feature patterns of speech, and identifying which of said stored feature patterns is most similar to said test pattern, said test patterns and feature patterns being divisible into corresponding frames of information, the improvement comprising: (a) input means for introducing a test pattern into said system; (b) memory means for storing said plurality of feature patterns; (c) judging means for sequentially accessing said stored feature patterns in said memory means and calculating similarity values between said corresponding frames of said test patterns and each of the feature patterns; (d) function generator means for developing a threshold similarity pattern having threshold similarity values for each of the said corresponding frames, said threshold similarity values having predetermined initial values; (e) comparator means for comparing the similarity values calculated by said judging means with the threshold similarity values of the threshold similarity pattern, and generating a stop signal when any similarity value of a given corresponding frame is less than a threshold similarity value for that frame; (f) control means responsive to said stop signal for precluding said judging means from calculating any further similarity values for the frames of the feature pattern then being judged and causing said judging means to access and calculate similarity values for another stored feature pattern in said memory means; and (g) identification means for identifying which of said feature patterns, the calculated similarity values of which as compared to the test pattern have no frames with similarity values less than said threshold values, is the most similar to said test pattern.