A Principal Component Analysis Approach to Noise Removal for Speech Denoising

Abstract

There is still remnant noise which affects the quality of speech data after traditional speech signal denoising for the speech data. In order to improve the effect of speech signal denoising, a speech denoising method based on the principal component analysis was proposed. Firstly? an embedded matrix that contains all information of collection signals is obtained by using dynamic embedded (Dynamic embedding, DE) technology in order to meet the needs of the principle of principal component analysis. Secondly, the principal components of speech signal are transformed through the principle of principal component analysis. By analyzing the principal components of speech signal, the low order principal components which associated with the big eigenvalues reflect the correlated speech signals could be selected to reconstruct the speech data. The required number of principal components is determined based on the Bayesian information criterion. Finally, the speech data are reconstructed by suitable number of the low-order components to remove the uncorrelated noise. In order to prove the effectiveness of the denoising algorithm, several experiments are conducted by the traditional filtering algorithm and the algorithm of this paper. All of the results indicate that the de-noising algorithm based on principal component analysis can achieve a better effect. And the signal waveform of the principal component denoising method is more full and more close to the original speech signal.