An audio scrambling degree measure based on information criteria

In this paper, we study the normalized measure of audio scrambling degree which is consistent with the human auditory system. Firstly, we study the average short term audio entropy property. Randomly selecting consecutive sample points from a given-length audio analysis frame, we compute its short term entropy and then its corresponding mean value. With the increasing length for an analysis frame, the average short term entropy increases as well and finally keeps stable around global entropy for whole audio sequence. During the process, for an updated audio with more chaos, its average short term entropy reaches a big value in an earlier stage and then keeps in this level. Therefore, we propose an audio scrambling degree measure based on multi-scale average short term entropy. Experimental results suggest the method can obtain excellent measuring results for different types of audio, such as speech and music.