Research Of Blind Source Separation For Speech Signal Based On Firefly Algorithm

The problem of blind source separation has always been a research focus in the field of digital signal processing. And by the joint efforts of the researchers over the years, a variety of effective algorithms have been proposed and successfully applied to image processing, speech enhancement, medical signal processing and other fields. The blind source separation algorithm is mainly divided into two parts: the independence criterion and the optimization algorithm. For the traditional blind source separation algorithm, the optimization algorithm is used to optimize the objective function which is composed of the independence criterion. The result of separation is usually affected by the initial value and the nonlinear function which lead to the poor robustness of the algorithm. In order to solve the above problems, some intelligent optimization algorithms are introduced into the blind source separation by the researchers, such as particle swarm optimization algorithm and the fish algorithm. The objective function is optimized by means of the survival of the fittest and good results have been achieved. As a new intelligent swarm optimization algorithm, the firefly algorithm has the characteristics of simple concept, simple process and less adjustment parameters, which has been applied in many fields. So we can try to introduce the firefly algorithm to the problem of blind source separation.According to the main features of the blind source separation algorithm and the firefly algorithm, the main work of this thesis is:(1) Firstly, the theoretical basis of blind source separation is systematically introduced, including the mathematical model, the hypothesis condition of the blind source separation and the analysis of the uncertainty of the separated signal. The traditional blind source separation method is used as the key point and several commonly used criteria of independence and blind source separation algorithms are researched. And their characteristics are analyzed and compared. Finally, the performance evaluation indexes of the blind source separation is also introduced.(2) The related theory of firefly algorithm is introduced in detail. The characteristics of the firefly algorithm are analyzed. A new blind source separation algorithm based on standard firefly algorithm is proposed, and this algorithm is applied to the blind source separation of speech signals. The similarity coefficient and waveform of the separated signal are analyzed by simulation experiment. The simulation results show that the performance of the proposed algorithm is better than the two classical algorithms, compared with the two typical blind source separation algorithms.(3) In view of the problem that the standard firefly algorithm is easy to fall into local extreme value and the convergence rate is slow, the performance of the blind source separation algorithm may be affected. In this thesis, the improved method of standard firefly algorithm is analyzed, and a new algorithm based on clustering algorithm is discussed. On the basis of this algorithm, a new blind source separation algorithm based on improved clustering algorithm is proposed, which uses Cauchy mutation strategy and neighbor strategy to improve the algorithm. Compared with the blind source separation algorithm based on standard firefly algorithm, the separation effect of the speech signal and the convergence rate of the algorithm are improved, so the effectiveness of the algorithm is proved.