Continuous speech recognition system

Claim 1

A continuous recognition system comprising: means for developing a similarity measure and a path information at each time point (i,j) to determine a matching path by which an input pattern corresponds with a continuous reference pattern in the order of monotonous increase of i and j based on the similarity measure and path information obtained up to said time point (i,j) by making use of Dynamic Programming, i and j designating time points at time axes of said input pattern and said continuous reference pattern, respectively, said input pattern being given in a time series of feature vectors for an input speech, said continuous reference pattern being given in a time series of feature vectors for connected combinations of a plurality of reference word patterns, said similarity measure being determined by an accumulated distance between said feature vectors of said input pattern and said reference word patterns, and said path information at said time point (i,j) indicating the number of time points where said matching path up to said time point (i,j) intersects said input pattern axis at points corresponding to an initial time point of one of said reference words: means for setting the minimum similarity measure obtained for a reference word as an initial value for a next subsequent reference word; means for checking whether or not said path information at said time point (i,j) is within a restriction range along said input pattern axis on either side of said time point (i,j), said restriction range being defined for each said time point (i,j) in accordance with at least said path information; means for excluding a time point whose path information is outside said restriction range from development of said similarity measure and path information; and means for deciding recognition results of said input pattern on the basis of the matching path showing the minimum similarity measure at the end time point of said input pattern.