Search

In this paper, a new method to detect the presence of double MP3 compression is proposed. The method is based on the Difference of Calibration Histogram (DCH) characteristic. The features are extracted from the difference of the statistics of MDCT coefficients between the test audio and its calibrated audio. The performance of the scheme is tested on a database containing approximately 13,830 audios, a support vector machine (SVM) is applied to the features for detecting the double MP3 compression. Experimental results show that the proposed features are sensitive to double MP3 compression and achieve higher accuracy than the existing state-of-the-art, especially for same-transcoded and down-transcoded MP3 recompression. The method can also be used to estimate the bit-rate of the first-time MP3 encoding from the recompressed MP3 files.