听觉掩蔽效应语音增强的改进算法

A new speech enhancement based on human auditory masking

含噪信号利用掩蔽效应去噪后,噪声估计的误差导致语音失真。在利用听觉阈值计算谱减系数时提出了一种改进的计算方法,通过增加修改参数来抑制语音的过分衰减,减少了语音失真,然后基于MMSE准则对增强的语音谱再进行平滑处理,进一步抑制音乐噪声。实验表明该算法在不影响语音失真的基础上,提高了信噪比,消除了音乐噪声,主观测听的语音音质明显提高。

After a noisy signal is denoised by auditory masking, the noise estimation errors lead to voice distortion. In the use of auditory masking value of the coefficient spectrum, an improved method of calculation is proposed and amended by adding parameters to curb excessive attenuation voice and reduce voice distortion, then based on the MMSE guidelines, to enhance the voice spectrum further smoothing treatment and further inhibit music noise. Experiments show that the algorithm improves signal to noise ratio and eliminates music noise without effects on voice distortion. The quality of subjective listening to voice improves markedly.