Method to cancel noise in speech based on matching pursuits algorithm in time-frequency domain

Abstract

As the development of the technology of speech signal processing, the robust of the technology becomes more important. In order to obtain the useful information, how to eliminate the noise in the real environments and how to recognize the speech in the noise environments become the important research. With the research of speech signal in time-frequency domain (TFD), a method is proposed to cancel the noise in speech, which is based on that the speech signal has been decomposed by matching pursuits algorithm. In this method, the signal should be decomposed firstly, and then the characteristic of the speech and noise will be showed clearly with using the sum of Wigner distribution of every decomposed part of signal as the Wigner distribution of the whole signal. In paper, the research has been carried out under the different SNR environments. From the experimental results, it can be seen that the novel approach is desirable in canceling noise, and is more effective than the other conventional algorithm to cancel the white noise in noisy speech signal. At the end of paper, the disadvantage of this method is described. (4 pages)