【摘要】 在基于时延估计的声源定位系统中,由于定位算法分为两个阶段:时延估计和定位,时延估计阶段的误差会在定位阶段被放大,导致声源定位的成功率和精度较低。从原始信号去噪,时延值插值和定位算法三方面入手,提高声源定位的精度。结合自行设计的四元十字麦克风阵列,给出一种新的时延值筛选算法,实现了一个室内声源定位系统。实验结果表明,在二维定位场景中,该系统对声源方位角的估计成功率超过70%,平均误差小于5°;该系统对声源距离估计的成功率和精度与声源方位角有关,当声源方位与X、Y轴的夹角不超过15°且声源距离不超过2.5 m时,声源距离估计的成功率能达到50%以上。更多还原

【Abstract】 In sound source localisation( SSL) systems based on time delay estimation( TDE),the localisation algorithm is divided into two stages: the time delay estimation and the sound source localisation. The success rate and precision of SSL could be very low because the estimation error in TDE stage would be magnified in SSL stage. Proceeding from 3 aspects: raw signal denoising,time delay value interpolation and SLL algorithm,we improved the precision of SSL. Combined with a self-designed four-element cross microphone array we presented a new time delay value screening algorithm,and realised an indoor SSL system. Experimental results showed that in scenario of 2-D SSL,the success rate of sound source azimuth estimation in the system was at least 70%,its mean error was less than 5°. The success rate and precision of sound source distance estimation were related to real sound source azimuth and distance. When real sound source distance was less than 2. 5 m,and the included angle between sound source azimuth and X-Y axis was no more than 15°,the success rate of sound source localisation could reach 50% or higher.更多还原