Weighted noise subtraction and adaptive soft-thresholding approach to speech enhancement

This paper presents an improved and efficient method of speech enhancement using non-linear weighted noise subtraction (NWNS) and empirical mode decomposition (EMD). In the first stage, a noise subtraction method is used for enhancement of noisy speech by adopting a non-linear subtractive factor. The factor is estimated by a two stage refinement of estimated degree of noise (EDON). This stage performs well for wideband stationary noises. Most of the real world noise is narrowband non-stationary and EMD is a powerful tool for analyzing non-linear and non-stationary signals like speech. EMD decomposes any signals into a finite number of band limited signals called intrinsic mode functions (IMFs). In the next stage, we apply EMD through a time domain adaptive thresholding strategy adapted to the IMFs. This stage performs well for narrow band non-stationary noises. We use white, pink and high frequency channel noises in order to show the performance of the proposed algorithm.