近场声源三维定位MUSIC算法研究

信号到达方向(DOA)估计是阵列信号处理的最主要目的之一,基于语音阵列的DOA估计与传统的窄带远场信号的DOA估计在信号模型和处理方法上有很大区别.本文把经典的MUSIC算法推广到语音阵列处理中,实现了声源三维定位.该算法的基本思想是:把宽带语音信号通过短时傅立叶变换转换为独立离散频率点,利用MUSIC算法求出每个/部分频率点的空间谱矩阵,然后加权平均这些空间谱矩阵.通过搜索该平均空间谱矩阵,找到信号源的位置.最后,在16元均匀圆环阵的情况下,选用双语音信号源进行了仿真.仿真结果表明,该算法定位性能优良.

Speech Source 3D Localization MUSIC Algorithm in Near Field

The signal direction of arrival (DOA) estimation is the primary task of the array signal processing. The signal models and processing methods are different between the DOA estimation based on speech array sensors and the classical parameter estimation based on narrow-band signal in far field. Three-dimension localization MUSIC method for speech array is presented using classical MUSIC algorithm. The wideband speech signal captured by the array is transformed into several discrete independent frequency bins using STFFT. The spatial spectral matrices of these frequency bins are calculated according to classical MUSIC algorithm. And then, the weighted average of these spatial spectral matrices can be obtained. The location of the speech source can be found by searching in the average spatial spectral matrix. Computer simulation was carried out using uniform circle array!with 16 sensors and double speech signal sources, and simulation results show that the proposed algorithm has good localization performance.