Continuous speech recognition apparatus

Claim 1

A continuous speech recognition apparatus comprising: an acoustic analyzer circuit for extracting feature parameter data of an input speech in each of frame; first memory means storing a plurality of reference pattern data each including reference parameter data of N frames; a partial similarity calculating circuit for calculating a partial similarity between the feature parameter data of each frame which is supplied from said acoustic analyzer circuit and each reference parameter data which is read out from said first memory means; second memory means for sequentially storing partial similarity data from said partial similarity calculating circuit for a predetermined number of frames; an operation circuit for calculating similarities between the feature pattern data including N feature parameter data of the input speech and the reference pattern data on the basis of the N partial similarity data which correspond to each of the reference pattern data and are present in at least one subperiod, and for selecting the largest one of the calculated similarity data; third memory means for storing the largest similarity data from said operation circuit, and reference pattern indication data and subperiod indication data which respectively indicate the reference pattern and the subperiod which are associated with the largest similarity data; and a recognition circuit for detecting a plurality of series of continuous subperiods during the speech interval, calculating the sum of similarity data associated with each of said plurality of series of continuous subperiods and recognizing the input speech on the basis of a series of reference pattern indication data corresponding to a series of contiuuous subperiods with which the largest one of the sums of similarity data is associated.