Speech Enhancement using Discrete Wavelet Transform and Nonlocal Means Estimation

V. Sagar Reddy and Dr.K. Bikshalu

In this paper, we propose a novel speech enhancement algorithm based on wavelet decomposition and Non-Local Means estimation. The NLM, a patch based denoising method is extensively used for two-dimensional signals like Image, but its use for one-dimensional signals is getting more attention recently. The NLM based approach is more effective in removing low-frequency noises based on nonlocal similarities present among the samples of a signal. But it suffers from the issue of under averaging in the high-frequency regions. The DWT is a more efficient technique for removing high-frequency components but it requires more decomposition levels in order to remove noise from the low-frequency regions. As speech is no stationary, so NLM alone is not effective to remove the noise components from speech signal unlike Image denoising. To address these issues first, we decompose the signal into low frequency and high-frequency regions as an approximation and detailed coefficients respectively using discrete wavelet transform. The approximation coefficient is processed through NLM estimation to eliminate the noise component from low-frequency regions. The high-frequency noise components are eliminated by thresholding the detail coefficients at each level. The simulation result shows that the proposed method gives better speech enhancement in terms of objective quality measures under various noise environments with different SNR levels.

Volume 11 | 04-Special Issue

Pages: 1834-1840