Speech Noise Reduction with Wavelet Transform Domain Adaptive Filters

Abstract

Adaptive filters are one of the most promising solutions to several signal enhancement problems in a non-stationary environment. However, depending on the characteristics of the signals and noise, the processing complexity and convergence speed for adaptive filters vary. Therefore, it is often preferred to apply adaptive filters in the transform domain to reduce complexity and increase convergence speed. In this paper, the application of the LMS (Least Mean Square) algorithm, which is the most preferred algorithm of adaptive filters in the field of speech noise cancellation, in the wavelet transform domain was studied. For this purpose, improving speech signals with different Signal to Noise Ratio (SNR) using Wavelet Transform Domain LMS (WTD-LMS) algorithm in the proposed method was applied. The results obtained were evaluated with measures that are frequently used in speech enhancement applications. It is observed that the success of the proposed method outperforms adaptive and traditional methods for two sensor measurements are available.