Performance test of parameters for speaker recognition system based on SVM-VQ

Many parameters can be extracted from a speech signal, including pitch, LPCC, ALPCC, P ARC OR, MFCC, AMFCC, RCEP etc. These parameters have different effectiveness for a speaker recognition system. In order to improve recognition efficiency and obtain a practical speaker recognition system, it is necessary for research feature parameters, that is the main contents of this article. Different parameters are extracted using the method of signal processing including time domain and frequency domain in this paper. These features are analyzed and compared, and the mixed features´ effect on the performance of the recognition system is also researched. In order to compare the efficiency of some parameters for speaker recognition system, the identification method based on SVM-VQ on time-frequency domain is chosen. Compared with SVM or VQ recognition method, the method based on SVM-VQ takes less computation, and has better noise immunity and better robustness. The experimental results show that some parameters have great influence on the system performance, such as pitch extracted using wavelet, LPCC and ALPCC, as well as MFCC and AMFCC. The experimental results also show that the recognition rate is obviously improved using mixed parameters in the system.