| In the real environment, the speech is often affected by interference from noise inthe external environment, making the speech quality decline. Especially in the case ofstrong background noise, speech may be completely submerged in noise, and can’tdistinguish. Speech de-noising has been an important branch in signal analysis andprocessing, its main target is extracted as much as possible the pure primitive voicefrom the speech signal with noise. Over the years, people have developed manymethods for the voice noise reduction. Most of the methods in a higher signal noiseratio (SNR), can better attenuation of the noise in the noised speech, but in a lower SNRthe de-noising performance is poor.Independent component analysis (ICA) is developed in recent years as a new kindof signal separation technology, the basic idea of this method is based on non-Gaussiansignal as the research object, under the premise of the assumption of independence willseparate the original signals from multi-channel observation signals. this method can bepreferably separated the implicit independent source signals from the multi-channelobservation signals. The emergence and development of the ICA provides a new ideasfor the method of speech noise reduction.This article in view of the actual environment of the speech with noise, ICAmethod based. First create a linear instantaneous mixing and linear convolution mixingsystem model, According to the theory of ICA have deduced a kind of time-domainindependent separation analysis algorithm, and through the computer simulation resultsverified the effectiveness of the algorithm. For the independent component analysismethod was applied to the actual environment of the voice mixed with noise, putforward a implementation method with the independent component analysis under thelinear convolution mixed model which closer to the voice and the noise mixed in theactual environment. Converted time-domain signals into frequency-domain signalsthrough the Fourier transform. And then use the frequency-domain FICA algorithm formixed signals’ separation, thus separated the noise and the speech signal from thenoised speech. Simulation results show the method has a better separation effect of theactual environment with noised speech. This paper also has put forward several kinds ofICA method under the single channel. Finally, we have designed speech de-noisingsystem on the DSP. |
