Development of Robot Audition with Two-Dimensional Resolution Exploiting Two-Channel Directional Sensors

通过计算机仿真，研究了基于加权维纳增益（WWG）的双指向性传声器波达方向（DOA）估计方法的性能。结果表明，该方法在多源情况下优于传统的广义互相关方法和两种高分辨率方法，具有上级的优越性。该结果已发表在IEEE Transactions on Speech，Audio and Language Processing上。接下来，我们对该方法在存在人头相关传递函数（HRTF）影响的情况下的DOA估计性能进行了评估。这是假设麦克风埋在机器人体内的情况。在评估之前，我们使用通过CT测量的年骨的三维形状和通过激光扫描仪测量的头部表面形状制作了对应于四个人的精确的四个头部模型。 ...更多信息 四人中有两人是男性，另外两人是女性。我们使用这些头部模型在消声室中测量了四组HRTF。作为评价结果，我们证实了我们的方法是优于传统的方法上级。但是对于多源环境，性能没有达到实用水平。然而，我们的方法在多个源环境中找到一个最大功率源时获得了非常高的性能。在IEICE的技术报告和IPSJ东北分支分部的研究会议上，我们发表了相关论文。为了提高多源环境下的性能，我们进一步提出了利用最大功率源的DOA估计来估计多个源的方向的迭代方法，并通过计算机仿真对该方法进行了评估。结果表明，该方法即使在多源条件下也能获得非常高的性能。上述研究结果表明，我们已经开发出了高性能的机器人听觉DOA估计功能。少

We investigated the performance of direction of arrival (DOA) estimation method based on weighted Wiener gain (WWG) utilizing two directional microphones as sound sensors through computer simulations. As the result, we confirmed that our method WWG is superior to the conventional methods of generalized cross-correlation (GCC) and two high-resolution methods particularly in the condition where multiple sources exist. This result was written in the paper and published in the international journal IEEE Transactions on Speech, Audio and Language Processing.Next, we evaluated the DOA estimation performance of our method in the case where effect of human head related transfer function (HRTF) exists. This is for the case when microphones are assumed to be buried in the robot's body. In advance to the evaluation, we made precise four head models corresponding four individual persons using three-dimensional shape of the year bone measured by CT and head surface shape measured by laser scanner. … More Two of the four are male persons and the other two are females. We measured the four sets of HRTF in an anechoic chamber using these head models. As the evaluation results, we confirmed that our method is superior to the conventional methods. But the performance did not reach the practical level for multiple source environments. Nevertheless, our method attained remarkably high performance in finding one source of maximum power in multiple source environments. We presented papers on this result in technical report of IEICE and in research meeting of Tohoku branch division of IPSJ.To improve performance in the multiple source environments, we further proposed iterative method to estimate directions of multiple sources by using the estimated DOA of the source with maximum power and evaluated this method through computer simulations. As the result, we confirmed that this method can attain remarkably high performance even in the multiple source condition. We presented papers on this result in technical report of IEICE.The above research results demonstrate that high performance DOA estimation function for robot's audition has been developed. Less