Abstract: Noise spectrum estimation algorithms in single channel speech enhancement approaches play an important role, in order to improve noise estimation and updating ability of noise spectrum estimation, this paper combines Minimum Estimate（MS）algorithm, and improves the recursive averaging parameter based on Improved Minima Controlled Recursive Averaging（IMCRA）noise spectrum estimation algorithm, at the same time, uses a recursive way to improve the minimum value of smoothing power spectrum. The spectral subtraction method is used for noisy speech signal denoising, from an objective and subjective, the performance of different algorithms is evaluated, and comparatively analyzes the speech segment Signal-to-Noise Ratio（segSNR）, PESQ scores, MOS scores under different noise environment and SNRs before and after. Experimental results show that the proposed methods can track the changes in noise signal and improve speech quality.

Key words: noise spectrum estimation, Minimum Estimate（MS）, Improved Minima Controlled Recursive Averaging（IMCRA）, spectrum subtraction, speech enhancement

摘要： 噪声谱估计算法在单通道语音增强方法中起着重要作用，为了改善噪声谱估计算法对噪声的估计和更新能力，结合最小统计（MS）算法，对改进的基于控制的递归平均（IMCRA）噪声谱估计算法的递归平均参数进行改进，并用一阶递归的方式对平滑功率谱的最小值进行改进。采用谱减法对含噪语音信号作去噪处理，从客观和主观两方面对不同算法的性能进行评价，对比分析不同噪声不同信噪比下增强前后语音的分段信噪比（segSNR）、PESQ得分、MOS得分。实验结果表明，提出的方法能够更好地跟踪噪声信号变化，改善语音质量。

关键词: 噪声谱估计, 最小统计算法（MS）, 改进的最小控制递归平均算法（IMCRA）, 谱减法, 语音增强