一种改进的基于DCT的语音增强算法

An Improved Speech Enhancement Algorithm Based on DCT

提出一种改进的基于离散余弦变换的语音增强算法。在信噪比较低时,传统的基于离散余弦变换的语音增强算法效果较好,能较大幅度地提高信号的信噪比;而当信噪比高时,利用这种方法会滤掉一些有用的信号成份。新算法首先计算出所有高阶离散余弦变换系数对应的时域信号中语音信号出现的可能性大小,然后根据某个阈值计算是否在估计噪声信号绝对值的均方差时保留该系数。实验结果表明在含噪语音信号的信噪比高于10dB时,新算法较传统的基于离散余弦变换的算法具有较好的性能。

An improved algorithm for speech enhancement based on DCT（Discrete Cosine Transform） is presented. In the condition of low SNR, the traditional speech enhancement algorithm based on DCT has good effects. The SNR can be improved effectively. In the condition of high SNR, some useful signal components may be filtered out using the traditional algorithms. Using the new algorithm, the statistical probability that the finest DCT coefficients belong to speech signal is computed first. Then according to the adaptive thresholding values, whether these coefficients are reserved in computer median absolute deviation of the finest DCT coefficients or not is determined. The simulating results show the superiority of the proposed method over a recently reported enhancement algorithm based on DCT.