Tri-Dimension Sound Localization by Binaural Model Using Self Organizing Maps
Tri-Dimension Sound Localization by Binaural Model Using Self Organizing Maps
复制标题
DOI：
10.1299/kikaic.74.2577
复制
发表时间：
2008-10
期刊：
影响因子：
--
通讯作者：
Zhong Zhang;I. Kazuaki;T. Miyake;T. Imamura;S. Horihata
中科院分区：
文献类型：
--
作者：
Zhong Zhang;I. Kazuaki;T. Miyake;T. Imamura;S. Horihata
众所周知，在声音定位系统中使用两个麦克风在应用于辅助听力受损者和类人机器人时具有许多优点。本文提出了一种基于双耳模型的声源定位方法，该方法利用了两个传声器观测信号的特征差异。该方法首先基于自组织映射（SOM）算法学习双耳观测特征功率谱之比和双耳声音到达时间之差所对应的特征向量，然后生成三维声源方向的SOM.将声源的特征向量输入到SOM中，并且通过在映射的参考向量空间中搜索获胜节点来估计作为与输入的特征向量最近的参考向量的节点，并且输出与获胜节点相对应的声源方向作为估计的声源方向。用五种不同方位的声音对该方法进行了实验验证，正确率达98.7%。关键词：声音定位，自组织映射，双耳模型，人机接口，仿人机器人
It is well known that there are many advantages in using two microphones in sound localization systems when applied to assist the hearing impaired and in humanoid robots. In this paper, we propose a novel sound localization method based on the binaural model, in which the feature differences of observed signals obtained from two microphones are used. In our method, �rst the feature vectors corresponding to the ratio of the observed signature power spectrum between both ears and the difference between the arrival times of the sound in both ears are learned based on the algorithm of Self Organizing Maps (SOM) and then the SOM of the 3-D sound source direction is created. The characteristic vector of the sound source is input into the SOM and the node that is the nearest reference vector to the input characteristic vector is estimated by searching for the winning node in the map's reference vector space, and the sound source direction corresponding to the winning node is output as the estimated sound source direction. Sounds fromve kinds of objects from many directions were used to experimentally conrm the effectiveness of the method and a correct answer rate of 98.7% was obtained. Keywords: Sound localization, Self organizing map, Binaural model, Human interface, Humanoid robots