Research And Application Of Feature Clustering Method Based On Music Melody

In the retrieval mechanism based on melody, the main research contents are melody feature extraction, feature express and feature matching engine,in the present study,melody matching algorithm and melody model high robustness modeling had made some progress, this paper is based on the content of music retrieval technology, aims to improve retrieval efficiency on the basis of retrieval accuracy.This paper involves speech signal processing, MIDI thematic extraction, pattern recognition, data mining and other areas of related technologies. The main contents include melody extraction of MIDI format music, feature extraction of hum waveform file, clustering of feature database and melody matching problem, etc. Arounding the research content launched the following research work:1. Collecting MIDI format music and analyzing their format information.of MIDI sound track music melody extraction. Considering extract melody feature as feature database data storage.This method provides data support for music library clustering analysis and retrieval matching.2. For reference group improvisations melody feature extraction method, and made supplement experiment base on time-domain analysis and method of frequency domain.Invocate extract characteristic data of Matlab environment to VS.This practice improved the original system operating performance,and provides practical stronger experimental environment for follow up study.3. In data mining technology,clustering method is often used for extensive data analysis research.To improve retrieval efficiency of the paper, put automatic grouping unilateral continuous matching clustering algorithm before database retrieval of packet unilateral continuous based on clustering algorithm for audio research and analysis of two level matching algorithm and the understanding.The experimental proofs of clustering algorithm could improve the searching efficiency in basis and ensure retrieval accuracy.4. Finally,combined with features clustering algorithm and linear matching algorithm,design a music retrieval system based on the melody testing, and through different test,make analysis and evaluation on experimental data,and verifies the performance advantages of clustering algorithm in this paper and retrieval algorithm. This paper focus on MIDI thematic extraction,melody features data clustering,mel-odies retrieval matching and hum retrieval system construction of the four main problems such as study of melody.Focus on the melodies information features clustering and melodic matching algorithms.Puts forward clustering algorithm with strong tolerance base on unilateral continuous matching.Realize the optimization of the candidate song collection.Test result shows that this method obviously decrease the match the actual time needed to search for large-scale music database, provides value data support of the hum retrieval and provides a new way of multimedia retrieval based on content.