An Improved Speech Denoising Algorithm Based on Adaptive Least Mean Square

Abstract

The excess mean squared errors is the major disadvantage in speech denoising algorithm based on least mean square, which increase linearly with the desired signal power. In order to improve the performance of the speech which exhibits in large power fluctuations, a new adaptive speech denoising algorithm based on adaptive least mean square is proposed, the improved speech denoising algorithm solve this question through minimizing the correlation between the error difference and input difference vector when the speech signal is powerful. According to the different noise environments, the proposed algorithm is simulated using different noise power levels. Simulation results show that the excess mean square error can improve about 20 dB in different noise environments based on the same complexity. It is benefit to the speech recognition process in communication system of voice.