掩蔽法减少谱减法去混响中的音乐噪声

针对谱减法去混响中出现的音乐噪声问题,研究了一种基于频率归一化的掩蔽后处理方法。首先利用谱减法对混响信号去除混响,其次对混响信号与去混响信号进行频率归一化,然后在频域将去混响信号和混响信号对应相减,据此计算出掩蔽因子,最后运用掩蔽因子对去混响信号进行处理来减少音乐噪声。仿真实验中分别利用语谱图和语音质量感知评价进行了评估,语谱图表明后处理语音信号中音乐噪声明显减少,语音质量感知评价表明采用后处理方法提高了语音信号0.55的评分。

Reducing musical noise in dereverberation of spectral subtraction based on masking method

Aiming at the problem of musical noise in dereverberation of spectral subtraction, a post processing masking method based on frequency normalization is researched. Firstly, the reverberation signal is processed by spectral subtraction to remove the reverberation. Secondly, the reverberation signal and the dereverberation signal are normalized in frequency. Then the dereverberation signal and the reverberation signal are subtracted correspondingly in frequency domain to get the mask factor. Finally, the mask factor is used to deal with the dereverberation signal in order to reduce the musical noise. In the simulation experiment, the spectrogram and the perceptual evaluation of speech quality are used to evaluate the masking method. The spectrogram verifying the musical noise is reduced obviously in the post processed speech signal, and the perceptual evaluation of speech quality shows that the post processing method improves the score of the speech signal 0.55.