噪声相关条件下多传声器时延多尺度估计

MULTISCALE TIME-DELAY ESTIMATION OF MULTI-MICROPHONES WITH CORRELATIVE NOISES

提出一种基于小波分解的多尺度时延估计方法,解决非稳态、有色噪声且通道间噪声存在相关性的条件下提高时延估计精度问题,并把这一理论应用于声阵列对声源的定位。利用小波良好的时间-频率分解性能将非稳态信号分解为多个稳态信号,同一尺度内进行互相关获得多尺度时延值,并证明了小波分解不会改变信号之间的固有时延值。针对不同尺度内时延值不一致、无法识别出正确时延值这一问题,提出用三传声器之间时延匹配关系消除噪声相关性,来识别正确时延,从而实现非稳态、噪声相关条件下的时延估计。最后通过正四方形声阵列试验检验了该方法,试验证明:该方法不但提高了识别率、细化和突出了相关函数的波峰,还提高了时延估计的稳定性和精度。

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.