一种基于语谱图分析的语音增强算法

针对传统谱减法会产生“音乐噪声”及在低噪比环境下降噪效果不理想的问题,提出一种基于语谱图分析的语音增强算法.所提方法先对带噪语音进行谱减处理,再通过语谱图分析对处理后的语音信号进行去小亮块处理,即对“音乐噪声”进行有效的削弱,最后通过设置阈值,滤除低灰度值的成份,即消除残留近似高斯噪声.实验结果表明,提出的算法可以有效地解决谱减法后残留的“音乐噪声”及低信噪比下谱减法残留噪声较大的问题.

A Speech Enhancement Algorithm Based on Speech Spectrogram

Aiming at the musical noise produced by classical spectral subtraction algorithm which doesn＇ t work well in low SNR, the algorithm of speech enhancement based on spectrogram is proposed. Therefore, the steps of the speech enhance- ment algorithm based on speech spectrogram are as follows： Firstly, enhance the original speech by spectral subtraction al- gorithm. Secondly, weaken the musical noise by eliminate the small light blocks through the spectrogram analysis. Finally, eliminate the residual approximate gauss noise by setting a threshold to filter low gray level components. The experimental re- suits show that the proposed algorithm can effectively solve the problem that large musical noise and other noises are re- mained after spectral subtraction in low SNR.