基于相似性和新奇度从音乐中提取代表性摘录  

Representative excerpts extraction of music based on similarity and novelty

作　　者：吕波[1] 李建彬[1] 贺苏宁[1] 燕继坤[1] 

机构地区：[1]信号盲处理国防科技重点实验室,四川成都610041

出　　处：《计算机应用》2007年第3期740-742,共3页journal of Computer Applications

摘　　要：提出了基于相似性和新奇度提取音乐代表性摘录的方法。最大化片段与整个作品的相似性来找到最相似的摘录,最大化片段的新奇度来找到意义最丰富的摘录,并把这两个参数结合起来寻找最具代表性的摘录,同时引入数学形态滤波对音频信号预处理,消除信号中的非主要分量。实验结果表明,该方法能够找到重要的最具代表性的摘录,并且对音乐源只作了很少的假设。The method for automatically producing representative excerpts of music based on similarity and novelty was proposed. The segment similarity was maximized to the entire work to find the most similar excerpt, and the novelty of the segment was maximized to find the most meaningful excerpt. The two measures were combined to find the most representative excerpt. At the same time, mathematical morphological filtering was used to preprocess audio-signal to eliminate incidental elements. Experimental results demonstrate that the method can find the most representative excerpts, just using very few assumptions about the source music.

关 键 词：自相似性 相似性矩阵 新奇度 代表性摘录 

分 类 号：TP37[自动化与计算机技术—计算机系统结构] TP391.42[自动化与计算机技术—计算机科学与技术]