Adaptive noise reduction algorithm based on gradient in wavelet feature domain

摘要

Conventional linear system adaptive filtering techniques have been widely used in adaptive noise reduction problems. However, because of the linearity of the operation, it is difficult to suppress the noise and keep the speech signal using linear filters when the spectrum of a signal is somewhat wideband and nonstationary. In order to slove the problem, gradient-based adaptive learning algorithms with the characteristics of infinitely differentiable are presented to seek the optimal solution for noise reduction. Test on several different speech databases, the experimental results also prove the effectiveness of the proposed method.