Application Of Matching Pursuit Algorithm In Speech Blind Source Separation

Blind source separation refers to getting process observation signal recovered when thesource signal and the transmission channel is unknown. The blind source separation of thespeech as an important branch in blind source separation technology is a hot spot in the area ofsignal processing. In practice, many of the signals are sparse, or which are sparse through thecorresponding transformation such as DCT, FFT, and Wavelet etc. So the sparse componentanalysis has gradually become a new research direction in the field of BSS.Domestic and foreign scholars have carried out in-depth research on the problem of blindsource separation. Blind signal separation has attracted the attention of many researchers to usethe signal’s sparse characteristics. This thesis is based on the matching pursuit algorithm overcomplete Gabor dictionary tracking sparse decomposition.Firstly make the mixed speech signals sparse. Sparse decomposition in signal analysis isvery important in the signal processing, such as signal recognition, signal compression and signaltransmission. It is a new method of signal decomposition, the matching pursuit algorithm is akind of the most commonly used method; it can be approximated rebuilt signals. MP algorithm isa kind of new method based on Wavelet transform and it has high time-frequency resolutionwithout the band division in the analysis of the signals. The use of a prior knowledge of thesignals to select the best atom is more objective. The application of signal sparse decompositionwith MP is extensive applications such as the time-frequency analysis, data compression andsignal restoration etc. Based on the decomposition the results show good performance.MP algorithm is a greedy algorithm and the complexity decomposition restricts to lookwidely. Domestic and foreign scholars have conducted a lot of researches to realize fast signalMP decomposition algorithms, such as using FFT transform.Then blind source separation uses adaptively learning algorithm with activation function inmixed signals by MP sparse to decompose signs. In the actual situation it is very difficult to havea prior knowledge before separation and then using the learning algorithm source signal Kurtisadaptive. This algorithm through simulation proves that the separation effectiveness. In the endthe results show a certain improvement on the traditional algorithm then confirm theeffectiveness of the algorithm.