中文摘要：

为提高低信噪比环境下的语音可懂度,提出了一种基于联合失真控制的子空间语音增强算法。由于误差信号中的语音失真和残余噪声分量不能被同时最小化,同时,由语音估计器引起的语音放大失真超过6.02 d B时会严重损害语音可懂度。为此分别对语音失真和残余噪声进行最小化处理,最小化时把语音放大失真控制在6.02 d B以下作为约束条件,通过求解两个约束最优化问题得到两个不同的估计器,再对这两个估计器进行加权求和,得到一种基于联合失真控制的语音估计器。实验结果表明,相比于传统的子空间增强方法,在低信噪比环境下所提出的算法能更有效提高增强后语音的可懂度。

英文摘要：

In order to improve speech intelligibility in low signal-to-noise ratio environment, a subspace speech enhancement algorithm combined with distortion control is proposed. Due to the facts that the components of speech distortion and residual noise in the error signal can not be simultaneously minimized and that the amplification distortion of speech in excess of 6.02 d B caused by speech estimator will seriously damage the speech intelligibility, the speech distortion and the residual noise are minimized respectively, and meanwhile the speech amplification distortion is kept below 6.02 d B as a constraint condition. By solving these two constraint optimization problems, two different estimators are obtained, and then a weighted sum of these two estimators is made to get the speech estimator based on combined distortion control. The results show that the proposed approach can improve enhanced speech intelligibility more effectively in low signal-to-noise ratio environment, compared with the traditional subspace enhancement method.