Abstract: Frequency domain subspace speech enhancement using fixed Lagrange multipliers to construct enhanced matrix, it can reduce the speech distortion and improve the speech intelligibility, but the major drawback is the residual musical noise. In the auditory masking, noise is masked by the strong frequency component near it in frequency domain. With the relationship between the frequency domain and the Eigen domain of signal subspace, the diagonal matrix of the gain function is calculated by using a variable Lagrange multiplier. Both comparative experiments and audition results show that the proposed algorithm can not only enhance the signal-to-noise ratio greatly, but also improve the subjective perception of voice quality significantly.

Key words: enhanced matrix, Karhunen-Loeve（KL） transform, Lagrange multiplier

摘要： 针对频域受限子空间语音增强在构造增强矩阵时，采用固定拉格朗日乘子，使得减小语音畸变和提高语音可懂度的过程中，有音乐噪声残留，提出一种变拉格朗日乘子的算法。利用听觉特性中较强的频率成分对噪声进行掩蔽，通过掩蔽阈值的频率域与子空间特征值之间的变换算法，用变量控制子空间拉格朗日乘子计算增益函数的对角矩阵。对比实验和试听结果表明，提出算法增强的语音信号不仅信噪比有较大提高，语音质量主观感知度也有明显改善。

关键词: 增强矩阵, 卡亨南-洛伊夫（KL）变换, 拉格朗日乘子