为了提高在噪声环境下的语音识别性能,提出一种融合信号级和特征参数级抗噪的抗噪算法.该算法首先对带噪语音用最小均方误差估计法进行语音增强,后端对原始的带噪语音运用自相关法,以有效抑制加性和卷积噪声.实验结果表明,该算法能有效提高系统在噪声环境下,特别是低信噪比情况下的识别率. In order to improve the performance of speech recognition in noisy environments, an anti-noise algorithm based on signal-level and feature-parameter-level anti-noise is proposed.This algorithm first performs speech enhancement on the noisy speech with minimum mean square error estimation, The autocorrelation method is used to effectively suppress the additive and convolutional noise.The experimental results show that this algorithm can effectively improve the recognition rate of the system in noisy environments, especially in the case of low signal-to-noise ratio.