A noise reduction method for speech signal combining ICA-R and EMD-wavelet
Yangyang Qi, Guofu Wang, Miao Yu · 2013
A novel noise reduction method combining ICA-R and EMD-wavelet is proposed in this paper. Because the single channel ICA is an extreme ill-condition problem in mathematics, which common ICA is incapable to solve, we exploited the Empirical mode decomposition (EMD) technique to expand the single channel received signal into several intrinsic mode functions (IMFs). The noise is suppressed through two steps, first, wavelet threshold denoising is applied to the frontal two IMFs, second, the signal and the remaining noise are separated by ICA-R. The reference signal is constructed by the low frequency IMFs, helping ICA-R to extract the object speech signal. Simulation results indicate the proposed method can recovery the speech signal from noisy signal effectively, especially when noise-to-signal ratio is high.