摘要： 在经典子空间语音增强算法中，因语音特征值估计偏差会造成语音失真和音乐噪声。针对该问题，提出一种联合听觉掩蔽效应的语音增强算法。该算法联合掩蔽阈值自适应调节噪声特征值的抑制系数，并利用维纳滤波对音乐噪声的抑制性，对该特征值并行修正，最终还原出纯净的语音。实验结果证明，该算法在白噪声和有色噪声的背景下，与经典子空间的语音增强算法相比，能提高信噪比，减少语音失真和音乐噪声。

关键词: 子空间, 维纳滤波, 特征值, 语音增强, 掩蔽阈值

Abstract: In the traditional subspace speech enhancement algorithm, the bias of the estimated speech eigenvalues leads to speech distortion and residual noise. A new algorithm is advanced to correct them. The algorithm adjusts adaptively coefficient ahead noise eigenvalues jointing masking effect, modifies concurrently the clean speech eigenvalues by wiener filter for its restraint to noise especially music noise, attains the clean speech. Experimental results show that the proposed algorithm can achieve the higher SNR than the traditional algorithm in the same Gaussian white noise and train colored noise, and reduce speech distortion and residual noise.

Key words: subspace, Wiener filter, eigenvalues, speech enhancement, masking threshold

中图分类号: 

• TP18