支持得分矩阵的音乐检索技术.pdf

Computer Engineering and Applications 计算机工程与应用 2017 ，53（5） 17 支持得分矩阵的音乐检索技术 王培培，王 斌 WANG Peipei, WANG Bin 东北大学 计算机科学与工程学院，沈阳 110004 School of Computer Science and Engineering, Northeastern University, Shenyang 110004, China WANG Peipei, WANG Bin. Fast and sensitive matching of music using position specific scoring matrices. Computer Engineering and Applications, 2017, 53 （5 ）：17-23. Abstract: As an important way of music retrieval, humming query has gained wide attention because of its effectiveness and convenience. This paper proposes a novel retrieval technique based on the scoring matrices of humming, which can provide fast retrieval for humming query. In the proposed technique, the music database and humming given by users are first partitioned according to natural pauses, and k-means clustering algorithm is adopted to compute pitch similarity. This paper sets specific scoring matrix according to clustering. Based on scoring matrix, this paper further proposes a brute force pattern matching algorithm, as well as two accelerated methods. The experimental results demonstrate both the efficiency and effectiveness of the retrieval method proposed in this paper, and the PLA algorithm has more effective humming music retrieval result. Key words: humming; scoring matrices; partition music retrieval; query by humming 摘 要：作为音乐检索的重要方式，哼唱检索由于其有效性和方便性，引起了广泛的关注。对此提出了一种新的基 于得分矩阵的音乐哼唱快速检索技术，可以实现哼唱音乐的快速检索。首先根据哼唱音乐特征，将音乐数据库和用 户提供的哼唱片段，按自然停顿方式划分音乐的语句，同时使用K-means 聚类算法对音乐的语句片段进行音高相似 性计算，并根据聚类情况提取出位置特异性得分矩阵。此外，基于得分矩阵提出NA 匹配算法和两种加速分段计分 方法，分别是顺序前瞻计分SLS 算法和置换矩阵前瞻计分PLA 算法。实验结果表明所提出的基于得分矩阵的音乐 检索技术能够快速有效地返回查询结果，同时PLA 算法具有更有效的哼唱音乐检索结果。 关键词：哼唱；得分矩阵；音乐检索；哼唱检索 文献标志码：A 中图分类号：TP391 doi ：10.3778/j.issn. 1002-8331.1608-0325 1 引言 非常熟悉，于是人们可以将旋律哼唱出来，通过麦克风 随着现代多媒体技术和网络技术的迅速发展，多媒 数字化输入给计算机，计算机就可以使用搜索引擎去寻 体信息的数据量急剧增多，声音媒体是其中非常重要的