Speech Blind Separation Algorithm And Application Research

Blind source separation is along with the development of speech processing technology and it is one of the development of speech signal processing field. Blind source separation is told simply separate wanted inside source signal from a set of observed signals that its advanced knowledge is unknown. So far, the research of blind source separation focuses on the independent component analysis and computational auditory scene analysis, The technology is widely used in various fields, such as speech enhancement, communication, speech recognition, radar detection, etc.The research of blind source separation in this paper is from three aspects First is independent component analysis method. This paper proposed improved algorithm based on maximum information. The algorithm cited a sign function. It is previous to the algorithm which cited sigmoid function.compared to former algorithm, the new method greatly improved the separation of clarity and speech intelligibility. In this paper, speech source separation algorithm based on the minimum gain using method of matrix eigen value to solve separation of voice, its advantage is that compared with the natural gradient algorithm and information maximization algorithm it can not only omit the iterative operation, but also improved the similarity coefficient and crosstalk error of the separation speech, and improved the separation efficiency. The research of algorithm based on computational auditory scene analysis in this basis design a system that its voice decomposition feature is the starting time and end time. Finally, proved the validity of this system through comparing signal-to-noise of speech signal ratio before and after separation. Speech enhancement algorithm is one of the most important contents which related to people's life. The paper proposed a new speech enhancement algorithm based on the knowledge independent component analysis (ICA). The algorithm realizes multi-channel by introducing reference signal. Finally the paper proves the effectiveness of the algorithm by experiment, and the enhanced performance evaluation indexes are still signal-to-noise ratio of the separation.