一种鲁棒的双耳声源方位角定位方法
2022,30(11):204-212
摘要：在噪声和混响的声学环境中，基于双耳时间差的声源方位角定位性能会严重降低。针对这个问题，提出了一种基于子带选择和DBSCAN的双耳声源定位算法，首先，采用 Gammatone 滤波器将双耳声源信号分解为若干个子带信号；其次，根据子带能量大小进行子带通道数压缩；然后，根据子带信噪比大小获取最优子带，降低无关子带干扰；接着将子带信号进行分帧，根据互相关算法获取峰值处的数据点；最后，引入DBSCAN算法消除噪声点的影响，获取最优数据点，从而根据ITD定位模型判断目标声源方位角，实验结果表明，该算法在复杂的声学环境中，相较于传统的互相关算法，可显著提高双耳声源方位角定位性能。
关键词：双耳声源定位；数据压缩；子带选择；互相关算法；DBSCAN
A Robust Binaural Sound Source Azimuth Location Method
Abstract：In noisy and reverberant acoustic environments, the performance of sound source azimuth localization based on binaural time difference is severely degraded. To solve this problem, a binaural sound source localization algorithm based on subband selection and DBSCAN is proposed. Firstly, the binaural sound source signal is decomposed into several subband signals by using Gammatone filter; secondly, the number of subband channels is compressed according to the subband energy; Sub-band interference; then the sub-band signal is divided into frames, and the data points at the peak are obtained according to the cross-correlation algorithm; finally, the DBSCAN algorithm is introduced to eliminate the influence of noise points and obtain the optimal data points, so as to determine the target sound source according to the ITD positioning model. Azimuth. The experimental results show that the algorithm can significantly improve the azimuth angle localization performance of binaural sound sources compared with the traditional cross-correlation algorithm in complex acoustic environments.
Key words：Binaural sound source localization; data compression; subband selection; cross-correlation algorithm ;DBSCAN
收稿日期：2022-04-03
基金项目：
