Research On Sound Source Localization Technology Based On Dual Mini Microphone Array

Array signal processing technology plays an important role in many applications such as audio and video conference,robot and speech recognition.Sound localization is widely recognized as an important part of the array signal processing technology.In the actual environment with strong noise and high reverberation,the robustness of commonly used sound source localization algorithm is poor.Anti-noise and anti-reverberation are the characteristics of the Steered Response Power with phase transform(SRP-PHAT)algorithm,however,this algorithm needs a great deal of computation time.In order to reduce the computational cost and improve the performance of the SRP-PHAT algorithm in the noise and high reverberation environment,the following work is mainly done:1.Several microphone arrays with higher frequency are introduced,and the dual mini microarray model is introduced.The real environment voice recording based on the dual microarray is carried out by using human voice reading.2.To solve the issue of poor performance of endpoint detection in low SNR,a new detection method is studied.In this method,modulation domain spectral subtraction is introduced to remove part of the noise of the signal,and the empirical mode decomposition of the signal is combined with the logarithmic energy divided by the primary and secondary peaks of the correlation function.Finally,the endpoint detection is carried out.The experimental results show that the detection accuracy is mostly above 90% under low SNR,and the detection of the algorithm is better.3.The weight of the generalized cross-correlation function(GCC)in the SRP-PHAT algorithm is improved.The smooth correlation transformation(SCOT)function and the phase transformation(PHAT)function are combined to form a new weight function.The SRP algorithm of the improved weighting function combined with BP network for double microarray sound source location is studied.In this method,the low-resolution search space with 5° as a span is divided into regions according to its theoretical signal reception time difference(TDOA),which reduces the search space by 99.5% compared with the SRP full search method.The centroid coordinates of each region are taken as representative coordinates,the TDOA of representative coordinates is taken as search space look-up table,and the improved weighted SRP power of the look-up table is used as the input feature of the network,and the position of the output sound source is estimated through the continuous learning of the network.The simulation results show that the algorithm has better positioning performance.The average accuracy of azimuth prediction is 97.8%,elevation estimation is 98.5%,distance prediction is 99.8%,and sound source location is96%.Compared with the SRP full search method,the proposed algorithm improves the positioning accuracy by 68% and reduces the positioning time by 99.7%.