Research on Method of Speech Recognition Basing on MVDR and ICA

This paper discusses modeling method of minimum variance distortionless response (MVDR),which compares with linear predictive method. The result is that minimum variance distortionless response filter can afford better original speech envelop. MVDR parameter picking-up method is combined with independent component analysis, which compares with traditional speech recognition, and data has been analyzed. MVDR parameter picking-up method can improve recognition ratio of speech recognition system, can increase mean recognition time; and speech recognition system with ICA character transformation is more robust.