利用状态空间方法对说话人进行语音跟踪时,观测方程的非线性会影响说话人位置的估计精度。该文将迭代滤波理论与中心差分卡尔曼滤波技术相结合,提出迭代的中心差分卡尔曼滤波方法,并应用于说话人跟踪系统。仿真实验结果表明,该文所提出的方法减少了系统线性化误差,增强了滤波算法的鲁棒性,提高了说话人跟踪精度。
In the state space method based speaker tracking system, the nonlinearity of the measurement function degrades the localizing accuracy of the speaker tracking method severely. The iterated central difference Kalman filter algorithm, which incorporates the iterated filtering theory and the Central Difference Kalman Filter (CDKF) method, is proposed to reduce linearization error. In comparison with traditional CDKF method, the proposed method has higher tracking accuracy, faster convergence speed and more robust stability. Simulation results demonstrate the effectiveness of the proposed method,