基于EMD与K-S检验的语音增强方法

In the paper, we apply the Empirical Mode Decomposition of Hilbert-Huang Transform (HHT) to decomposing the noise-speech signal and the reference model noise respectively and get the intrinsic mode functions (IMF) of them, follow we sue the K-S test to calculate the similarity probability between the IMFs of noise speech and reference model noise, and then pick out the IMFs that come from the noise of noise-signal based on the similarity probability, so that we can eliminate the noise IMFs of noise-signal and reconstruct the speech signal using the other IMFs ,at last we get the purpose of speech enhancement. The experimentation result show that, compared with the speech enhancement method of wavelet, this method extremely eliminates the noise component of the noise-signal by suing the method and achieves a good speech enhancement effect, it has wide application prospect.