基于两级变参数滤波时延校正的声源定向

Two stage variable parameter filter time-delay correction approach to acoustic source direction

  • 摘要: 在人机交互、视频会议、军事国防等众多领域中,基于麦克风阵列的声源定位系统已经受到了人们的广泛关注。在空间噪声和混响的环境中,时延估计往往会出现很大的偏差,而声源角度的计算又依赖于时延估计的准确性。因此,本文在双阵元模型下提出了一种用于声源定向的时延校正算法,该算法采用一个权值系数可变的两级滤波器对GCC-PHAT-ργ估计出的时延点进行校正,有效地让偏差较大的时延点回归到正确的时延群当中,从而避免了错误时延值对进一步角度定向的影响。实验结果表明,本文提出的算法在不删除时延异常点的前提下,在误差均值和均方根误差等方面都表现出了很好的性能,而且对每组数据进行角度定向的时间都很短。理论分析与实验结果表明该方法不仅有效的提高了角度估计的估计精度,而且当说话人以正常走路的速度移动时,该算法可对说话人的方向进行及时的跟踪。

     

    Abstract: In the filed of human-robot interaction,Video conference and Military defense,the sound source direction system based on microphone array has already been getting increasing attention.Time-delay estimation point can show a large deviation when both spatial noise and reverberation exist,while the calculation of sound source angle relays on the accurance of time-delay estimation.Therefore,In this paper,we reported on time-delay correction approach used for acoustic source direction based on double microphones model.a two stage variable weighting coefficient filter is used to correct the time-delay point estimated by GCC-PHAT-ργ in algorithm,effectively making bigger deviation time-delay point return to the correct time delay group, thus avoiding impairing further angle estimation caused by wrong time-delay. As the results show,the proposed approach shows a very good perfomance in error average and RMSE without the decrease of sample qualities.It takes less time to locate the positon of angle for each data. Theory analysis and simulation results are provided to show that the proposed algorithm not only improve the estimation precision of angle,but also can track timely the direction of acoustic source when the speaker is moving at normal speed.

     

/

返回文章
返回