The Research Of Music Mood Classification Algorithm In Digital Audio System

With the development of the communication, computer and internet, large numbers ofinformation such as image, video and audio increase in a way of indexation. In the other hand,streaming media develops so quickly that people begin to face and handle much multimediacontent and can obtain rich music resources conveniently and fast. Therefore, it is the urgentneed to invent a new technique to manage the music resources effectively. The automaticclassification of audio signal has become the focus of engineering and academic.In the present, people can search music by choosing the name of the song or the singer, butwith the increasing development of living standard and the continuous enrichment of mentalfield, people hope to pick different types of music automatically according to the mood ofthemselves. To classify music according to the mood it shows has to be re-structured andoptimized based on the traditional way of classification, because it involves the comprehensivecharacterization and fuzzy classification of music signal’s natural attribute and social attribute.This paper presents a method of classifying music signal automatically according topeople’s mood. It selects those audio feature which can express the mood of music on thebasis of summarizing the results of previous research. Moreover, by discussing methods ofconstructing week classifiers based on Adaboost algorithm and GMM modal, it uses a kind ofmultilayer classifier architecture.The main work includes the following four aspects:(1) Analyzing parameters of music features and selecting available related characteristics ofmusic;(2) Constructing week classifiers based on Adaboost algorithm and GMM modal;(3) Establishing and training a strong multilayer classifier;(4) Presenting a audio classified system based on GMM modal and Adaboost algorithm andputting forward a method of designing week classifier group that can play the properties ofAdaboost algorithm, through combining the excellent robustness of GMM and the strongadaptability of Adaboost.