One speaker recognition method based on feature fusion

摘要

Now the most serious problem in speaker recognition is the robustness of the system. A new feature fusion method based on MFCC and bispectrum feature is proposed in this paper to improve the robustness of the recognition system. Focusing on the high dimension and the large amount of data among the bispectrum feature spaces, the 1½ -dimension (1½ -D) spectrum is selected in order to improve system efficiency. Finally, experiments are carried out based on TIMIT speech database. Comparing simulation results with MFCC proves that the algorithm can indeed enhance the robustness and the right recognition rate especially in low SNR. The right recognition rate increase by 12% in the case of 100 individuals with SNR of 10dB.