Audio Authentication Based On Music Content Analysis

With the rapid development of multimedia and network technology, it has become quite easy and convenient to copy, modify or transfer various digital audio data. At the same time, however, the authenticity of audio signals is being severely challenged. In an effort to enjoy more secure multimedia applications, there has been a strong need to protect the authenticity and integrity of audio content. Existing audio authentication algorithms mainly focus on either human speech or general audio with music as part of the test data, while dedicated research on music authentication has been somewhat neglected.In this paper, we concentrate on music content authentication which is aimed at protecting the integrity and authenticity of music signals, a special type of audio data. We introduce the background and representative application scenarios of audio content authentication, clarify the difference between traditional and media authentication, explain relevant concepts in the research, summarize most state-of-the-art audio content authentication algorithms in the literature and then point out some major problems. To overcome the weakness of previous methods, we propose two novel music content authentication algorithms.In the first algorithm, each music track is partitioned into a sequence of musical notes which correspond to minimum meaningful semantic elements of the music, thus settling the synchronization problem that remains a big challenge in most existing audio authentication algorithms. Chroma-based mid-level audio feature which can appropriately characterize the music content is adopted to generate the robust hashes. Three newly defined measures that characterize statistical and temporal properties of the distortion between the original and the dubious music are exploited and further combined with fuzzy classification to generate an authenticity degree, which complies with the inherent uncertainty of authentication.In the second algorithm, music signals are segmented into beat based frames, which not only endows the authentication units more semantic meaning but also perfectly resolves the challenging synchronization problem. The beat times of the music are extracted through an existing beat tracking approach which utilizes dynamic programming. Compared with the first algorithm, longer frames are used in chroma calculation so as to capture music information more precisely and at the same time not seriously affect system's robustness. A two-layer scrambling encryption procedure is integrated to relieve the threat from malicious block-wise vector quantization attack and further increase the security of authentication system. Finally, if authentication fails, three common malicious operations, i.e. cropping, adding and replacing are further classified by observing singular points in the beat alignment plot.Experimental results show that both methods feature high distinction between acceptable manipulations and malicious ones, achieve precise tamper localization and relatively accurate tamper classification.