A Robust Approach For Compressed-domain Audio Fingerprint

With the increasing development of multimedia technology, data compression and Internet technology, vast amounts of multimedia information already exists in our daily lives. It is becoming more and more difficult to quickly and accurately find the target audio (i.e. the audio file which we need) from a large number of audio files by the key information (such as title, author, date of publication, keywords, etc.) is mastered by seeker in advance. The method of audio fingerprint indexing which is based on the content, do not need key information, entirely depends on the audio content itself, and the result of candidates are very less, it is gradually playing an important role in the field of audio retrieval, audio recognition, the integrity checking for audio content. The current research mainly focuses on the theory and technology of indexing for non-compressed-domain audio fingerprint, few of studies are relevant to compressed-domain audio fingerprint, and rarely involve the robustness testing and improving of fingerprint.At first, this paper introduces the research background and significance of the audio fingerprint indexing technology, from a few points of view such as audio fingerprint diversity and robustness, compressed domain audio fingerprint index and fast audio fingerprinting algorithm to review the present research status, and then we conclude the related principles of compression-domain audio fingerprint.From the point of view of practical application, we propose a compressed-domain audio fingerprint algorithm on the basis of analysis and research about the compressed domain audio features, the algorithm is based on the MDCT spectral energy from the decompression process to directly calculate the compressed-domain audio fingerprint, there is no need to decode fully for compressed-domain audio. Considering the audio files collected by current handheld audio device are generally non-compressed audio, so this paper also designs a uncompressed audio fingerprint algorithm, this algorithm refers to standard audio compression procedure, it calculates the MDCT spectrum from the PCM audio signal, and then computes the fingerprint. The fingerprint generated by these two algorithms can be used to effectively retrieve in a same audio fingerprint database. Some of the technical features, for example the similarity, distinction, bit error rate and robustness of the fingerprint algorithm, are tested and analyzed, experimental results show that the algorithm has good performance, and the correct recognition rate meets the requirement of practical application. The paper also introduces a fast fingerprint indexing algorithm based on Hash model from the perspective of indexing and matching.On the base of analysis of the robustness of the test results, this paper researches a improved algorithm for the robustness of resisting a common time domain distortion phenomenon—linear speed change(LSC). which is optimized by introducing the shift invariance of the correlation function and the scale invariance of Fourier-Mellin transform. The test results show that the optimized algorithm based on the correlation function will improve the ability to against LSC from±5%to±7%, and this capacity of the another method based on Fourier-Mellin transform will increase to±10%. Two improved algorithms don’t affect the robustness of other common time-frequency-domain distortion.Finally, we designs a simple audio fingerprint identification system according to the proposed fingerprint algorithm, and system performance and technical indicators are tested and summarized.