Audio post-processing detection and identification based on audio features

As an important communication medium, audios are easily modified or tampered during transmission; thus the authenticity of audios is of high importance. This paper mainly introduces a method to detect audio post-processing based on audio features; the Support Vector Machine (SVM) is applied for classification during the detection. In the proposed method, the Mel Frequency Cepstral Coefficient (MFCC) and the Linear Prediction Coding (LPC) of host audios are calculated as audio features, to which SVM is applied to judge the authenticity of the audios. Experimental results show that the proposed audio feature based method can not only verify the authenticity of speech audio, but also have a significant effect on detecting different types of post-processing operations.