Research On The Content-Based Audio Retrival

The popularity of the Internet, the ubiquity of high-storage devices, the proliferation of Peer2Peer networks and world-wide low-latency networks has dramatically increased digital audio growth and access. But meanwhile, it is time-consuming and error-prone to manage collections of audio assets, which direct access to the rapid development of digital audio fingerprinting. Digital audio fingerprinting is a robust content-based compact signature that summarizes an audio recording. It is typically used for music copyright enforcement, automatic music identification and audio verification.This thesis introduces the audio fingerprint technology background, basic concepts and general framework. The key technique of digital audio fingerprinting are researched, and the main works are as follows:1) This thesis summarizes most state of the art audio fingerprinting algorithms and conducts an in-depth study of two effective extraction algorithms based on STFT.2) For the Philips audio fingerprinting algorithm has less noise robustness in the real environment, according to this point of view, we prove that to calculate the difference in power spectrum is equal to remove the additive in the power spectrum domain, and then two power spectrum based on time-frequency filtering of the audio fingerprinting extraction algorithm are presented. The improved algorithms are simulated and analyzed subsequently in real environment. Our experimental results show that the proposed audio fingerprinting algorithm improves noise robustness.3) After analyzing the present audio fingerprinting retrieval algorithms and the characteristics of the audio fingerprint database, a vector indexing method called locality sensitive hashing is proposed and the parameters are optimized. This method can effectively prune super large searching space and reduce time complexity of similar search process. The experiment results show that this method is still effective when the data scale is very large, and it has superior scalability than traditional tree-structured indexing methods.4) According to the algorithms and theories researched in this paper, an audio fingerprinting retrieval system is designed and implemented. The retrieval mechanism combined with time continuity of the audio fingerprinting is proposed. The experimental results verify that the system is effective and efficient.