哼唱搜索中一种基于DTW的旋律相似度算法  

Melody Similarity Algorithm Based on DTW Applied in Query by Humming System

作　　者：叶晓勇[1] 刘夙[1] 张贻雄[1] 郑灵翔[1] 

机构地区：[1]厦门大学通信工程系,福建厦门361005

出　　处：《电声技术》2013年第4期36-38,43,共4页Audio Engineering

摘　　要：哼唱搜索由于其用户要求低,应用方便,易于普及的特点,是当前音乐搜索应用中的主流方式,而旋律的相似度计算是哼唱搜索中直接影响搜索排名结果的关键技术。动态时间规整作为度量时间序列相似性的经典方法,是旋律相似度计算的主要方法之一。根据哼唱搜索的特点,提出了一种基于DTW的旋律相似度算法,实验结果表明,该算法进一步提高了哼唱搜索结果的精准度。Query by humming is the most popular method used in music research applications for its features of user-friendly and easy to popularize. Melody similarity calculation is the key technology in query by humming which would decide the result of retrieval directly. Dynamic time warping , as a classic method used for time based on sequence similarity measuring, is also one of the most important methods applied in melody similarity calculation. A melody similarity algorithm based on DTW is pro- vided. The experimental results show that great improvements have been achieved by this algorithm in accuracy of the result of query by humming.

关 键 词：哼唱搜索 旋律相似度计算 动态时间规整 

分 类 号：TN912[电子电信—通信与信息系统]