Study Of Speech Enhancement Algorithms Based On Microphone Array

Voice communication is an important means of human social interaction. It is the oldest,the most convenient, and the most effective methods of information exchange in humandevelopment history. However, the size of environmental noise and the quality oftransmission medium have seriously affected the quality of voice communication. In theprocess of handling noise, the noise and signal that the traditional single microphone receiveare often overlap in the time and spectral, and compared to the conventional method, usingthe microphone array(Microphone Array，MA)is a important way to improve the quality ofvoice communications. The methods of microphone speech enhancement which uses the arraysignal processing techniques can fully use the spatial and temporal information of speechsignals. So, the method has a high spatial resolution, a flexible control of the beam, a highersignal gain and a stronger anti-interference ability and other characteristics. In conclusion, thestudy about algorithms and techniques of MA which has a good ability to de-noising for avariety of noise has a very important significance in the current field of the voicecommunication.The theory and technology of microphone array speech enhancement is firstly described inthis paper, and more, compared several classic speech enhancement algorithm with the MA.The thesis focuses on the algorithm of denoising delay estimation techniques, beam formingand two phases of denoising at microphone array speech enhancement process, and achievesan improved algorithm. In reverberation environment, extraction of no reverb voice signal,then use the generalized mutual coherence delay estimation method, the delay value error issmaller. And it overcomes the impact of Reverb for TDE well. In beam-forming anddenoising stage, the adaptive algorithm, based on the generalized side-lobe canceller (GSC),has good ability to suppress coherent noise, and requires the number of microphones biggerthan that of noise sources. Post filtering algorithm based on Wiener filter, has good ability tosuppress the incoherent noise, but has relatively weaker effect for the coherent noise. Theabove two algorithms have both advantages and disadvantages, and try to combine the twokinds of algorithms, thus the method to suppress the incoherent and coherent noises founded.The simulation of the improved algorithm using MATLAB software, has shown that thealgorithm has good ability to adapt the noise environment and to denoise. The test resultsshow that, compared with the traditional GSC algorithm and post filtering algorithm, theimproved algorithm has better reliability in complicated noise environment, but also hasoutput higher signal-to-noise ratio.