一种基于分频谱减的去混响方法

A Speech Dereverberation Method Based on Sub-frequency Bands Spectral Subtraction

混响是封闭空间内产生的一种声学现象,过度的混响严重影响通信质量。在谱减法去混响的基础上提出分频带谱减去混响的方法,基于每个频段上混响时间不同这一特性,将谱减法用到每个频率子带中,提高谱减法去混响的精度。分别采用语谱图和语音质量感知评价的评价方式对整体谱减和分频谱减的两种算法效果作对比,发现分频谱减效果更为显著。

Reverberation is a phenomenon in the field of room acoustics,and excessive reverberation seriously impacts on auditory sense. Because decay rates of reverberation on each frequency bands is different,the algorithm of spectral subtraction needs work on each frequency bands to improve the result of dereverberation. A new algorithm about spectral subtraction on sub-frequency bands is introduced for dereverberation. The comparison between dereverberation using spectrum subtraction on the whole signal and dereverberation using spectrum subtraction on the each sub-frequency bands is performed. Results using spectrogram and perceptual evaluation of speech quality measure show that the performance of dereverberation using spectrum subtraction on the each sub-frequency bands is better.