Abstract: This contribution presents an algorithm for speech enhancement by means of an improved spectral subtraction.In this algorithm，a noise spectrum estimation approach by minima controlled recursive averaging is utilized firstly，thus no voice activity detection is needed here，secondly，the occurrence possibility of “musical noise” is reduced by making use of a recursively calculated over-subtraction factor based on sub-band signal-to-noise ratio.Analysis and experiment results show that this algorithm eliminates the “musical noise” to a certain extent and effectively improves the output signal-to-noise ratio.

Key words: noise spectrum estimation, spectral subtraction, speech enhancement

摘要： 提出了一种基于改进的谱减法的语音增强算法。该算法首先利用了一种由最小值控制的递归平均的噪声谱估计算法，因而无需语音端点检测，其次利用一种通过递归计算得到的基于子带信噪比的过减因子，减小了产生“音乐噪声”的可能性。分析和实验表明，提出的算法对“音乐噪声”起到了一定的抑制效果，并有效地提高了输出信噪比。

关键词: 噪声谱估计, 谱减法, 语音增强