A stepgain optimization method by the background experimental scheme for an ANC SRS

This paper describes a stepgain optimization method of an adaptive noise canceller (ANC) using Normalized Least Mean Squares (NLMS) algorithm for a speech recognition system (SRS). In this proposed method, a stepgain is optimized by the results of the noise reduction performance analysis that is presumed from data outside of speech intervals. In a classical method, the result of the optimized stepgain was a little away compared with the strict optimal value because of the Fuzzy Controller. In this proposed method, the optimal stepgain is experimentally evaluated by the analytical evaluation modeling in the background scheme when the system is not busy. Therefore the obtained value of the optimal stepgain is corresponding to the strict value. In this paper, this approach is named the Background Experimental Scheme. Results of experiments using data in a running car (50[km/h]) are shown to prove effectiveness of this proposed method. In this proposed method, the gap between the obtained value and the strict value became about 75% less than a classical method that used the Fuzzy Controller (SN-Ratio 5[dB]).