Parameters for Speaker Identi cation

Eigen-MLLR coe cients are proposed as new feature parameters for speaker-identi cation in this paper. By performing principle component analysis on MLLR parameters among training speakers, the eigen-MLLR coe cients (EMCs) are derived as the coe cients for the eigenvectors. The discriminating function of the new EMC features based on the Fisher criterion is found to be ten times larger than that of mel-frequency cepstral coe cient (MFCC) features, for distinguishing speakers. The speaker-identi cation accuracy using the EMC features are shown to be signi cantly better than that using MFCC features, especially when the quantity of enrollment data is limited. It is also shown that properly combining MFCC and EMC features can achieve a signi cant error rate reduction on the order of 50%-60% as compared to using MFCC features alone.