Abstract: According to the situation that precision and efficiency of search can not achieve desired effect based-on music retrieval，presents a kind of hierarchical clustering algorithm based-on unilateral continuous matching，which is clustering the character of music before matching them.According to the similarity the algorithm clusters the character space and forms different kinds of clusters.Then it marks on the cluster center，and has a more precise matching from the database which has been clustered.Experiments show that the algorithm not only increases the efficiency of the music match，but also ensures a rather high accuracy.

Key words: feature extraction, unilateral continuous matching, K-means, Linear Alignment Matching（LAM）, fault tolerance

摘要： 针对基于内容的音乐检索在查准率和查找效率上不能同时达到理想效果的情况，提出一种基于单侧连续匹配的分层聚类算法，即在匹配之前对音乐特征进行聚类。根据相似度值对特征空间进行分类，形成不同簇类，并对簇中心进行标记，再对聚类后的数据库进行精确匹配。实验表明，该方法不仅提高了音乐检索的效率，而且保证了较高的检索准确率。

关键词: 特征提取, 单侧连续匹配, K-means聚类, 线性对齐匹配（LAM）, 容错性