对LMS自适应算法进行了详细的性能分析与讨论，针对LMS算法运算较复杂、适应性较弱、稳定性差的缺点提出了一种HLMS（混合LMS）算法。建立了自适应噪声抵消系统，利用MATILAB软件对食堂、体育馆两处的录音信号进行计算机语音去噪仿真分析。实验结果表明，两种自适应方法均能有效抑制各种噪声污染，提高语音信噪比为60%~85%，而且对引入的语音信号失真也较小，二者均在10%以下。HLMS算法比LMS算法更简洁，LMS算法调节性能敏感于参考输入的方差，而HLMS算法敏感于参考输入的均方根值，因此HLMS算法的适应性比LMS算法更强，在非平稳随机输入情况下HLMS算法更能提供稳定的工作性能，更能较好的恢复原始语音信号。Abstract: This article analysed and discussed the performance of the LSM adaptive algorithm. Aiming at the LMS algorithm?蒺s shortcomings， such as complex algorithm， weak adaptability， and poor stability etc， it proposed a HLMS （Hybrid LMS） algorithm， established an adaptive noise canceling system， and had the speech denoising simulation of two voice recording signal on the canteens， stadium using MATILAB software by computer. Experimental results show that the two adaptive methods can effectively suppress a variety of noise pollution， and improve the voice signal to noise ratio of 60% to 85%. And the introduction of the voice signal distortion is also smaller， both of them are below 10%. HLMS algorithm is more concise than LMS algorithm. LMS algorithm’s adjusting performance is sensitive to the variance of the reference input， while HLMS algorithm is sensitive to the RMS value of the reference input. So the adaptation of HLMS algorithm is better than the LMS algorithm. In case of non-stationary random input， the HLMS algorithm can provide more stable performance and better recovery of the original speech signal.