Speaker recognition system capable of accurately selecting inhibiting reference patterns by using small amount of calculation

Abstract From the Canadian patent

In a speaker recognition system, a tree-structured reference pattern memorizing unit has first through M-th node stages each of which has nodes and each of which memorizes a reference pattern of inhibiting speakers the , reference, pattern of each of the nodes of (N-1)-th node stage representing acoustic features, in the reference patterns of the predetermined ones of the nodes of N-th node stage. An analysis unit analyzes an input verification utterance and converts the, input verification utterance into feature vectors. A similarities calculating unit, calculates similarities between the feature vectors and the reference patterns of all, of the inhibiting speakers. An inhibiting …