Study On Sound Source Orientation Estimation Algorithms Based On Microphone Array

The signal which is collected by traditional single microphone,limited in the acquisition range of voice signal and easily affected by the factors of environment noise and room reverberaton,has larger distortion.With the development of multimedia conference,hearing aid and other products in array technology,the voice signal processing technology is not only used for collecting the voice signal which is in expectation,but also needed to locate it accurately. However,these requirements can not be satisfied by the traditional single microphone voice signal processing technology.The feasibility of realizing sound source orientation estimation based on the microphone array has been verified in this thesis with using microphone array from two aspects:algorithm theory and computer simulation.During the three kind of methods for sound source localization, the method based on time delay estimation and spatial spectrum is focused on in this thesis.The performance of various algorithms mentioned in this thesis is also verified in aspects of estimation precision, convergence speed and computation complexity through a series of computer simulations and comparative analysis of experiments.Firstly,the simulation analysis of different kind of weighted functions which are common in generalized cross correlation under ideal conditions is presented in this thesis. Aiming at the disadvantage that PHAT weighting function is susceptible to the factors of environmental noise and room reverberation,higher estimation precision can be achieved in the low SNR and reverberant conditions by improving the structure of PHAT weighting function.Secondly,the algorithm of least mean square adaptive time delay estimation is studied and the nonlinear function between step factor and steady state error to form variable step size LMS algorithm on the basis of LMS algorithm is established,which eliminate the influence of uncorrelated noise and overcomes the contradiction between step factor and steady state error.Thirdly,the algorithms of MUSIC, MUSIC roots and ESPRIT are researched on algorithm theory,estimation precision and resolution with simulation contrast.The problem of mirror orientation of sound source in space is solved by designing and optimizing the array structure about to eliminate the inherent defect of plane array.Then,several coherent algorithms are also studied in this thesis to achieve the angle estimation of spatial sound source aiming at the problem that MUSIC algorithm is unable to estimate the coherent source signals.Finally,the DOA estimation of spatial sound source is realized in this thesis by combining with TDOA and spatial spectral search.With this method,not only the disadvantage of secondary accumulation in time delay estimation error can be avoided,but also the shortcoming of extensive calculation exist in the of method spatial spectral search can be overcomed.