The Study Of Blind Speech Source Separation In Noising Environment

Blind Source Separation (BSS) has been a new method in the research area of signal processing. BSS is a powerful tool to tackle those problems because it can reconstruct to original signals from the observed signals without prior knowledge of the mixing system and source signals. Now, BSS has developed rapidly, and it has been used in many fields such as wireless communication, the procession of array signals, speech separation, image procession, biological medical, seismic detection, radar and sonar, voice eliminator, and so on. The mixture of speech signal separation is an important aspect in BSS, it's also a challenging problem in signal processing.In this thesis,the fundamental theories of BSS and main methods for speech separation are exploited and investigated. First of all, blind separation algorithms of linearly instantaneous mixture system mixed source signals are studied. There are three main algorithm are introduced——Informax, FastICA and JADE. At present they are used in many areas of signal processing. Then blind separation algorithms of linearly convolutive mixture system mixed source signals are studied. The algorithm of blind source separation of this kind of separation molde includes two algorithms: The first kind is in the time domain separation algorithm, another kind is in the frequency domain separation algorithm.At present, there are many problems in speech signals blind separation, denoising and multichannel blind separation; it's worth us to discussing. The solution of these questions will bring us significant value on the examination of sound signal. This article mainly studies the application on blind source separation of instantaneous liner mixture model and convolutive mixture model by using ICA. It also studies ICA usage in the signal denoising under the background music and the multichannel blind separation.Through speech separation experiments, using natural environment speech signals, applying blind source separation algorithms mentioned in chapter two, we make a computer simulation to separate the mixed speech signals in the real environment and present the results of simulation tests. This article proved the effectiveness of the algorithm. And compared the separated performance of the various algorithms.