MULTISCALE TIME-DELAY ESTIMATION OF MULTI-MICROPHONES WITH CORRELATIVE NOISES
Sun Shu-xue · Zhendong yu chongji · 2008
In order to improve the accuracy of time-delay estimation(TDE)when the signals are non-stationary with colored noise or the noises of some different channels are correlative,a new method for multiscale TDE based on wavelet decomposition was proposed,and applied to source localization with the aid of acoustic array.The non-stationary signals can be decomposed into several stationary signals by fine time-frequency characteristics of wavelet decomposition.Multiscale time-delay value can be achieved by use of cross-correlation of stationary signals decomposed by wavelet at the same scale and it is proved the intrinsic time-delay values of signals in different channels will maintain unchanged after wavelet decomposition.When the signals are decomposed at different scales,time-delay values are different,so that the right time-delay value can't be correctly identified.In order to solve this problem,the method for eliminating the correlativity of noises by matching time-delays of three microphones was proposed.The method is validated by the experiment with pyramid acoustic array composed of four microphones.The results present that the new method improves recognition rate,makes the correlation peak fine and obvious,and improves the stability and accuracy of TDE.