中文摘要：

针对DCT系数的非高斯性以及广义高斯分布和α稳定分布在音频DCT系数建模上的局限性,提出了一种混合模型.该模型由广义高斯分布和α稳定分布线性加权而成,其加权系数由遗传算法产生.模型的准确性由衡量2个概率分布之间差异的物理量Kullback-LeiblerDivergence（KL散度）测量.实验结果表明,提出的混合模型接近于真实分布,并可用于音频检索和隐写分析.

英文摘要：

Considering the non-gaussian property of discrete cosine transform（DCT） coefficients of audio and the limitation of modeling the DCT coefficient distribution with a single distribution such as generalized gaussian distribution（GGD） or alpha stable distribution,we propose a mixed model.This mixed model is a linearly weighted average of GGD and alpha stable distribution,while the weight values are obtained using a genetic algorithm.The model accuracy is measured with the Kullback-Leibler Divergence,a metric for evaluating the difference between two probability distributions.Experiment results show that the mixed model is close to the true distribution of DCT coefficients and is of practical use in the field of audio retrieval and steganalysis.