Research On Speech Signal's Blind Separation Using Kalman Filter To Denoise

BSS(Blind source separation,BSS)is an unsolved problem of signal processing which only rise in 1990's.It means the process of extracting or recovering sources from the observed mixed-signals without any prior knowledge of sources or the channel.Due to that,it is widely applied in the fields which couldn't have a good solution if the traditional signal processing methods are used,such as biomedicine,radar and communication,data mining,speech signal analysis and image analysis,meteorological analysis,seismic exploration,etc.Because source signals may consist of various noises more or less in the practical application,it will extend the range of the application if BSS can work well under the noisy environment.Based on this situation,the thesis did some research on speech signal's blind separation under the noisy environment.The thesis's main method was that:the signals were filtered before BSS,for the purpose of enhancing the speech signals and weakening the noises,the filter was a Kalman filter.Then Jade algorithm,Fastica algorithm were used to separate the mixed-signals.The two algorithms are relatively mature algorithms which are used widely and acknowledged.After that,a GUI(graphical user interfaces) with Matlab was designed to perform the process directly.At last,robust features were extracted from the unmixed-signals,then the thesis compared the two algorithms's separation performance according to the results.The thesis did some accurate research on speech signal's blind separation under noisy environment by means of designing and simulating the system,which would be very useful to the late application and research of BSS.