Study On Internet Intelligent Music Retrieval Technologys Based On Adaptive Segmentation And SVM Algorithms

As the network has become an important communication channel for music, music intelligent search has become a hot topic of the Internet. How to find needed music in the vast music library quickly and effectively is a very meaningful research topic. However, due to the use of music on the Internet is mostly text information retrieval system as the main basis for the search, it is difficult to meet the needs of intelligent retrieval problems; the same time, such as google, baidu and other commercial search engines are also facing time-consuming and associated text information retrieval, effort and one-sided issues. In addition, the terminology, characteristics of key words identifying, the underlying characteristics and high-level semantics of the "semantic gap" and other aspects of the complexity, it's an urgent need to find a more effective music retrieval technology for future intelligent Internet applications providing a good platform support.While traditional methods of music retrieval research has been theoretically studied and recognized, but there are a number of their own shortcomings and deficiencies, especially in the era of the Internet to pursue a more intelligent. So, more intelligent technologies are required, such as data mining technology. Therefore, this paper retain the traditional retrieval methods, and propose a new music retrieval method combined with a segmentation technique based on adaptive feature recognition/extraction and SVM classification of content feedback technology.The paper, by consulting relevant literatures, with the Internet music features as the starting point, analyzed some typical music theory knowledge about retrieval technology, application model and existing problems. This paper studied the use of data mining technology in the intelligent application of music retrieval, mainly proposed a new method of adaptive segmentation to identify/extract the feature key method based on other sub-word methods, and the introduction of support vector machine based on the relevant feedback technology. Therefore, the key elements of this thesis are divided into two parts:Adaptive segmentation of Internet music feature recognition/extraction and classification method based on support vector machine intelligence search technology for Internet music analysis. The latter part of the first part is to prepare, then they can be introduced from two aspects. First, the use of single-class SVM to solve the relevance feedback in the process of retrieving music semantic gap problem, and then through a variety of the SVM classifier selection strategy to improve retrieval performance. Experiments proved that the method to obtain better accuracy, and thus more practical.About the adaptive characteristics of technology in music recognition/ extraction, the traditional segmentation algorithm relied heavily on dictionaries and statistics, so some unstructured datas and professional new words can not be effectively identified, especially for highly specialized field. This paper is based on "2-gram" to achieve a statistical model which can well adapt to the corpus of information segmentation algorithm, and the experiments prove that both time and accuracy can meet of the Internet intelligent music retrieval implicational needs.Based on key words in the music feature extraction, the paper goes into the music retrieval relevance feedback research with support vector machine. Firstly, with the operating principles of SVM, this paper leads the traditional single-class support vector machines relevance feedback method, the most point is to put on an improved way to make up for the shortcomings of traditional methods, and through libsvm classified toolbox experiments using the both styles of music database and emotional database to test the validity of the method, the average classification accuracy comes to 52%. Meanwhile, the paper also further diversified the selection based on support vector machine to retrieve the application of intelligent music. Describes how to construct the initial sample set and a new distance-based variety of the selection method and on this based on a variety of the SVM-based music retrieval model for intelligent Internet. Ultimately, by matlab and libsvm for music retrieval experiments, the results show t- hat the proposed algorithm compared to the literature of feedback method showed better performance.