Study On Spatial Processing Techniques Of Speech Enhancement

Speech signal processing plays an important role in modern information society. But speech signals are often disturbed by various interferences, such as background noise and speech from other speakers. Besides, the room reflection and environment reverberation are other factors that affect the speech quality. All of these not only descend the validity of all kinds of speech signals processing methods, but also deflate their applicable areas. As a result, speech enhancement is indispensable in speech signal processing. Because of the non-stationary of speech signal, variety of the noise interference and complexity of environment reverberation, speech enhancement is a challengeable and attractive research area. Due to the limitation of the quantity and property of single microphone, we always can not obtain satisfied enhanced speech signal by single microphone. Then speech enhancement based on microphone array became a hot research field since 1970s. With the development of advanced signal processing theory, especial the great progress of array signal processing and non-stationary signal processing technology, speech signals can be enhanced using more and more new algorithms with microphone array. So the purpose of the thesis is to explore more effective speech enhancement algorithms with microphone array based on spatial processing techniques for better performance and more comprehensive applicable areas.The contents of this thesis are as follows:First, the characteristics of speech signal and the basic theory on speech signal processing are summarized. Besides, the history and development of speech enhancement with microphone array are briefly reviewed.Second, the generalized sidelobe cancel (GSC) algorithm is carefully studied. For the disadvantage of traditional GSC, the traditional GSC algorithm was improved with a view to its structure and its algorithm. As a result, the GSC-like algorithm has a better performance and wider applicable areas.Third, by exploring blind signal separation (BSS), some new speech enhancement algorithms which do not need prior knowledge about speech source are proposed in this thesis.Finally, some computer simulations show us that the improved algorithms proposed by the thesis worked out satisfactorily.