基于内容的音乐检索研究

对音乐的特征表示进行了详细的阐述，详细介绍了乐曲相似性度量方法，并字符串编码方法对音乐特征进行编码，然后使用蛋白质序列局部比对方法对2239首中国乐曲和3960首西方乐曲的音乐库进行大量的检索实验。获取了大量的实验数据，对实验结果进行分析，并取得了较好的检索性能。最后针对音乐检索过程中的每次比对都相互无关的特点，具有可以很好地进行并行化特性，对算法进行并行化，实现在4台高档微机构成的群集上进行了并行检索实验，其检索时间约为串行检索时间的1／4，有很高的加速比。

Survey of music retrieval based on content

The representation of music feature is stated and the measure method of music similarity at some length is introduced, and the music feature is encoded by using string encoded method. And then, using the local comparative sequence method of protein sequence, from the music database which has 2 339 Chinese music and 3 960 west music, it conducted a great number of retrieve experiments, and attained amount of experiment data, the experimental results are analyzed, good retrieve performances is achieved. In the end, aimed at the characteristic of music retrieving which is irrelevant when comparing substrings and can be easily parallelized, it parallelized the retrieve algorithm, and carried a parallel retrieve experiment on a cluster which is made of 4 top grade PCs. The parallel retrieve time is accordingly a quarter of sequence one, and has achieved high acceleration rate.