【摘要】 针对传统的可控波束声源跟踪方法运算量大,难以实时处理的缺点,提出了一种基于卡尔曼滤波和预测的可控波束多声源跟踪方案,集成了语音活动检测,为运动声源分别建立状态模型和观测模型.卡尔曼预测为下一步跟踪提供了基准,免除了传统可控波束方法的全局空间搜索,可以有效地减少运算量,同时解决了无声段声源轨迹的重构问题.卡尔曼滤波可以抑制突发性的噪声,提高定位精度.仿真实验结果表明文中提出的方法的可行性.更多还原

【Abstract】 Traditional acoustic source tracking using steered beamformer is highly computational demanding,thus can hardly be applied in real-time processing.In this paper,we propose a multiple acoustic source tracking method using steered beamfomer based on kalman filtering and estimation,which integrate voice activity detection,and establish status model and evaluation model for each moving source.kalman estimation provide a baseline for next tracking,make it unnecessary to do a global searching which efficiently cut computational amount and solve the problem of tracking when the source is shortly silent.kalman filtering can repress sudden noise,raise the precision of tracking.The simulation result shows that our method is feasible.更多还原