METHOD AND SYSTEM FOR GENERATING SUPPRESSION STANDARD PATTERN OF SPEAKER RECOGNITION, NAMELY, COHORT AND SPEAKER COLLATING DEVICE INCLUDING THE SAME SYSTEM

摘要

PROBLEM TO BE SOLVED: To provide a method for generating a suppression standard pattern with uses adaptation and a speaker collating device. ;SOLUTION: An input speech is sent to an analysis part 10 and converted into feature vectors by the analysis part 10. The feature vectors analyzed by the analysis part 10 are sent to a suppression speaker selection part 40 and a feature vector storage part 30. The suppression speaker selection part 40 calculates the similarity between a standard pattern stored in the suppression speaker standard pattern storage part 20 and the feature vector of an object person sent from a feature vector storage part 30 to select a suppression speaker. The standard pattern of the selected suppression speaker is sent to an estimation part 50. The estimation part 50 finds a mapping function from the feature vector space of the suppression speaker to the feature vector space of the object person based on the standard pattern sent from the suppression speaker selection part and the feature vector of the object person sent from the feature vector storage part 30. The found mapping function is sent to an adaptation part 60. The adaptation part 60 uses the mapping function sent from the estimation part 50 to adapt the suppression standard pattern selected by the suppression speaker selection part 40. The adapted suppression standard pattern is an output.;COPYRIGHT: (C)1998,JPO