NL MS(normalized least mean square)算法是是自适应信号处理中应用最广泛的算法之一 ,它的最大缺点是收敛速度慢 ,对非平稳信号自适应能力差。为提高声回声抵消算法的收敛速度 ,采取了将语音分析中常用的线性预测作为前置滤波器的方法来减少参考信号的相关性 ,采用步长控制的方法来提高收敛速度 ,同时保持了 NL MS算法稳定性好和算法简单的优点。最后在模拟环境下对算法进行了测试 ,证明了比 NL MS算法具有更高的收敛速度和回声抵消量 The NLMS (normalized least mean square) algorithm is one of the most widely used algorithms in adaptive signal processing. Its biggest drawback is its slow convergence rate and its poor adaptability to non-stationary signals. In order to improve the convergence speed of the acoustic echo cancellation algorithm, the linear prediction method commonly used in speech analysis is taken as the pre-filter method to reduce the correlation of the reference signal and the step control method is used to improve the convergence rate while maintaining the NL MS algorithm has the advantages of good stability and simple algorithm. Finally, the algorithm is tested in the simulated environment, which proves that it has higher convergence speed and echo cancellation than the NL MS algorithm