METHOD AND APPARATUS FOR VERIFICATION OF SPEAKER BY MIXTURE DECOMPOSITION AND IDENTIFICATION

摘要

PROBLEM TO BE SOLVED: To realize voice recognition with high precision. SOLUTION: A new device for using a new speaker verification method and MDD called mixture disassembling recognition (MDD) is shown. MDD fetches mixture component score information from a speaker individual recognition device and transmits this information while this information is still disassembled as a mixture of component score showing HMM state response before it is incorporated into a parameter of a single speaker individual recognition device. MDD3101 to 310N is very effective to improve the performance of the existing verification method based on a speaker dependent HMM group- normalized. MDD3101 to 310N is used together with the speaker dependent linear recognition device. This has a relatively small calculation and storage capacities. Thereby, the high performance of a hybrid MDD/CNHMM system can be achieved only by expanding a calculation and data storage asset to the minimum.