基于直接判决估计和预测估计的语音增强算法

Speech enhancement based on Decision-Directed incorporating predicted estimation

摘要:

对噪声环境下语音增强算法中的先验信噪比估计提出了一种新的估计算法。该算法结合了直接判决估计和预测估计方法，与传统的基于直接判决的先验信噪比估计算法相比较，当后验信噪比较大时，该算法减少了语音失真；当后验信噪比较小时，算法较明显地减少了音乐噪声。通过研究发现传统的直接判决方法就是该算法的一个特例，该算法可以通过一个权衡因子退化为传统的直接判决方法。实验结果也证实所提出的算法的优越性能。

Abstract:

A novel approach combining Decision Directed(DD) estimation and Predicted(PD) estimation was proposed to improve the priori Signal to Noise Ratio(SNR) estimation for speech enhancement in noisy environments. PD estimation incorporated the soft-decision scheme and speech presence uncertainty. In contrast to the conventional DD-based scheme, a less speech distortion was achieved when a posteriori SNR was higher; while a further reduction of musical noise was achieved when a posteriori SNR was lower. It was also found that a special case of the proposed method had been degenerated to the DD estimation with a weighting factor. Experimental results demonstrated the improved performance of the proposed algorithm.