Abstract: In order to improve the accuracy of noise estimation in the non-stationary environment, this paper proposes a new noise estimation algorithm based on constrained variance. The smoothing parameter can be calculated through constraining the variance. The noise power spectrum is estimated. Simulation results show that in comparison with other three methods, this algorithm possesses a more short delay in tracking the behavior of the background noise, and it has a lower mean square error of noise estimation. It is especially remarkable in the non-stationary noise and burst noise environment.

Key words: constrained variance, noise spectrum estimation, speech enhancement

摘要： 为了进一步提高非平稳环境下噪声估计的准确性，提出了一种基于约束方差的噪声谱估计算法，通过约束方差计算得出平滑参数，对噪声功率谱进行估计。实验结果表明，相对于其他三种算法，该算法能较低时延地跟踪背景噪声的轨迹，且它的噪声谱估计均方误差较小，在非平稳噪声及噪声突变环境下尤为明显。

关键词: 约束方差, 噪声谱估计, 语音增强