Content-based Intelligent Retrieval And Repetitive Detection Of Massive Audio

With the development of Internet data and the improvement of computingpower, we faced with the "information ocean" era. Especially for unstructureddata’s storage, such as audio signals, more and more demand for its storageand management is growing. Traditional text-based audio retrieval technologyprefers to use keywords related to audio content as query input text, then findout the complete audio information we are concerned. It needs human to textannotations of the meaning of the query audio, which leads those descriptionsubjectivity and incomplete, so we cannot accurately describe the richsemantic information and sensory information contained in the audio, it alsocan’t meet the rapid and efficient retrieval requirements under the massiveaudio environment. Content-based audio retrieval means detailed analysisthrough audio features, such as rhythm, intonation, loudness of audio data,then made that different audio data have different semantic query input, andthe same semantic audio to be similar in auditory perception. By doing this,not only avoids many disadvantages of text-based audio retrieval, also canadapt to the mass of retrieving audio environment demands, so content-basedaudio retrieval has become a lot of research emphasis and hotspot of scholarsboth at home and abroad.Based on summarizing some domestic and foreign scholars on the basis of research results, the relevant content-based audio retrieval technology arefocused on discuss in this paper. Based on Philips fingerprint extractiontechnology, emphasis on the characteristics of the audio classification andindexing technology is studied, then the audio similarity measurement methodto carry on the reasonable choice. Finally, combining Filter-and-Refineframe model on the basis of improved audio retrieval accuracy, this methodspeed up the retrieval speed with a higher margin of filtering, and be able tosupport real-time retrieval.The subject finally achieved retrieval audio clips from article130000audiolibrary within one second in the general configuration of PC, and the retrieverecall rate reached95%, the retrieval accuracy reached above97%. It has agreat significance for audio data access services under the age of the Internet.In this paper, the main work and research results are as follows:This paper elaboration basic digital characteristic of audio signal in brieffirst, mainly illustrates the features which we used to classify audios. Thenselect the Philips fingerprint extraction technology as the research basis,combining with the ABV algorithm was designed and implemented forcompression and classification of fingerprint. Finally filter large unrelatedaudio clips in a short period of time which reducing the matching calculation.Then FNV hash algorithm was improved in this paper, hybrid hashalgorithm is used to extract the hash index of Philips fingerprint. The hashtable we create is established, and high-efficiency for the distribution of the original audio fingerprints. Also, its conflict resistance is strong with a highcomputational efficiency.Combining Filter-and-Refine framework we implements the three layers offiltering audio retrieval system, accelerate the filtering speed and improve thematching efficiency.