Research On Source Localization Of Microphone Array In Indoor Environment

The sound source localization technology based on the microphone array has wide application prospects in the industrial and civil fields.The positioning research in the indoor office environment has certain research significance because of its application in the fields of intelligent monitoring and speech recognition.Therefore,this thesis studies the improvement of sound source localization algorithm and its actual localization performance in an indoor office environment.Regardless of algorithms simulation or the actual positioning application,it is necessary to determine the number of sound sources in advance,and then complete the positioning estimation of the sound source position.Under a single sound source,traditional positioning algorithms perform well in an ideal simulation environment,but in actual applications,due to poor anti-noise reverberation performance,the error is higher and the positioning performance is reduced.Under multiple sound sources,in addition to environmental noise and reverberation that will reduce positioning performance,mutual interference between sound source signals is also an important factor affecting positioning performance.Therefore,this thesis first uses the gerschgorin disks estimation to identify the number of speakers in the room.Aiming at a single person speaking indoors,a generalized cross correlation ameliorated phat algorithm(GCCAPHAT)joint noise masking function is proposed.This algorithm optimizes the phase weighting and combines the noise masking function to improve the estimation performance by sharpening the peak of the cross-correlation function.For many people speaking in the room,an improved algorithm based on delay sum beamforming combined frequency point selection and weighted covariance matrix(modify delay and sum beamforming,MDSB)is proposed.It first uses frequency point selection to eliminate the impact of grating lobes generated during broadband beamforming,and then uses a phase weighting function to process the data covariance matrix to further improve the algorithm's positioning performance.Finally,in the actual indoor environment,the real voice signals of the sound source acquisition platform are used to verify the effectiveness of the proposed improved algorithm.The main work of this thesis is as follows:(1)The theoretical basis of microphone array positioning technology.This thesis introduces the preprocessing methods commonly used in the location algorithm and different signal models in the process of sound source signal propagation,and briefly introduces the near and far field model of the microphone array and its array topology.Three kinds of classical sound source localization techniques are described and analyzed.Finally,the location technology based on generalized cross-correlation time delay estimation and delay sum beamforming technology are selected as the main research basis of this thesis.(2)There are two criteria for identifying the number of sound sources: Akaike information theory and minimum description length criterion,and explain their advantages and disadvantages.The principle of identifying the number of sources of the gerschgorin disks estimation(GDE)is analyzed,and the performance is estimated through comparative analysis of simulation and actual experiments to determine the GDE adjustment factor.(3)In the case where the number of sound sources is judged as a single source,the principle of generalized cross-correlation algorithm and several commonly used weighting functions are introduced.Aiming at the problem of low sound source localization accuracy in real environments,this thesis proposes the GCCAPHAT algorithm to suppress the noise signal frequency components and enhance the speech signal spectrum weight,and then use the idea of effective frame averaging to smooth the delay estimation function.In many experiments,compared with GCC-PHAT and MCPSP algorithms,the results show that GCCAPHAT algorithm has better robustness than the other two algorithms in different reverberation and noise environments.(4)In the case of discriminating the number of sound sources as multiple sources,the principle of the delay sum beamforming(DSB)algorithm and the speech broadband beamforming model are introduced.In this thesis,the influence of the side-lobe produced by beamforming on the location results of broadband speech in multi-source localization is discussed,and a modified delay sum beamforming(MDSB)joint frequency selection and weighted covariance matrix algorithm is proposed.This method uses the frequency of the voice signal and the spacing of the microphones to improve the frequency selection in the sub-band of the broadband beamformer,which greatly reduces the total computational complexity of the algorithm and eliminates the interference effects of side lobes.Afterwards,the data covariance matrix is processed with a phase weighting function to further improve the robustness of the algorithm.By comparing the original DSB and smooth DSB algorithms through experiments,the results show that the MDSB algorithm has stable positioning performance and better robustness.(5)Build a sound source acquisition platform for the actual indoor environment,and verify the positioning performance of the actual sound source through analysis of real voice.The results show that the GCC-APHAT and MDSB algorithms have good positioning performance in a real office conference room.