Statistical modeling of DCT coefficients for audio

Based on the sharp-peak and heavy-tail of statistical distribution of discrete cosine transform (DCT) coefficients for audio, generalized Gaussian distribution (GGD) and alpha-stable distribution are usually employed as modeling tools. In this paper, the Kullback-Leibler Divergence is applied to measure the difference between modeling result and the true distribution, and the experiment results show that compared with alpha-stable distribution, the generalized Gaussian distribution is closer to true distribution of DCT coefficients of audio, which proves that generalized Gaussian distribution could be utilized in the fields of audio retrieval and steganalysis.