Research On Sound Source Localization Based On Speech Enhancement In Complex Environment

Sound source localization is an important technology involving human-computer interaction,speech enhancement,and digital signal processing,which has broad application prospects.In practical applications,the presence of noise and reverberation often leads to a decrease in the performance of localization.Therefore,it is of great theoretical significance and practical engineering value to have research on sound source localization technology in complex environments.Firstly,the simulation and analysis of common generalized cross-correlation(GCC)time delay estimation method is performed in the thesis.The common GCC time delay estimation algorithm has poor time delay estimation performance in a relatively noisy environment,and it can not obtain accurate time delay estimation value.Therefore,in the thesis,speech signals received by microphone are processed with noise reduction.The commonly used noise removal algorithm from speech signals is the speech enhancement algorithm.This thesis studies and compares the commonly used speech enhancement algorithms,which shows that LMS adaptive filtering algorithm has a strong ability to remove noise.Therefore,considering the adaptive denoising of the speech signal is first performed before the generalized cross-correlation of the signal,then a time delay estimation algorithm based on LMS-PHAT is proposed.The simulation results show that the proposed algorithm can obtain high-accuracy time delay estimates under high noise intensity.Secondly,in the actual environment,especially in a closed indoor environment,reverberation is unavoidable,so the delay estimation of speech signals under reverberation conditions is considered.In view of this,this thesis analyzes the spectral subtraction dereverberation proposed by K.Lebart and JM.Boucher,and proposes an improved spectral subtraction dereverberation.The new algorithm is to compute the average for power spectral density estimate and the reverb power spectral density estimate to obtain the gain factor.Then,the spectral subtraction magnitude spectrum is obtained by multiplying the gain factor and the average amplitude spectrum.Finally,obtaining the speech signal of dereveration according to the IFFT transform of the signal formed by combining the spectral subtraction magnitude spectrum with the phase spectrum.The simulation results show that the improved spectral subtraction dereverberation has a better reverberation suppression effect.Therefore,by suppressing the reverberation signal and combining the LMS-PHAT time delay estimation algorithm,the accuracy of the time delay estimation value can be improved.Finally,the geometric model of the five-element microphone array is designed in the thesis,and a sound source localization system is constructed.The localization system is mainly composed of five parts,which are the de-reverberation processing of the sound source signal and the de-noising of the de-reverberation signal processing,generalized cross-correlation delay estimation,position estimation of source signals,and acquisition of target distance and azimuth.The localization system performs experiments and analysis on the localization results of different sound source position.The experimental results show that the localizaiton accuracy of the system is higher in the indoor environment.