Research on Speech Enhancement based on Improved Wavelet Threshold Selection and Least Mean Square Error Adaptive Noise Cancellation

Abstract

Aiming at the problem of speech enhancement disturbed by high-frequency and broadband noise, a speech denoising algorithm based on improved wavelet threshold and least mean square (LMS) erroradaptive noise cancellation is proposed. Firstly, LMS adaptive noise canceller is used to cancel some noise to obtain a speech signal with high signal-to-noise ratio, and then the wavelet analysis is used fordenoising and reconstructing to obtain the denoised speech signals. Matlab simulation experiments show that the improvedalgorithm is superior thanthe single algorithm. The output signal-to-noise ratio, visual effectand root mean square error are greatly improved.