基于分句匹配的哼唱音乐检索系统的算法研究与实现

SENTENCE-BASED HUMMING RETRIEVAL

提出一种以分句匹配为基础的哼唱音乐检索的系统架构及实现方法.首先建立以分句为基础的歌曲旋律特征库,之后采用改进的自相关函数法提取哼唱的基音序列,经规整化处理后,生成与歌曲旋律特征库所一致的特征,然后采用分句匹配方法进行检索.经过实验分析,结果表明这种新的处理方法对于哼唱音乐检索的正确率有明显的提高.

This paper proposes a humming music retrieval framework and new methods to implement it based on sentence. First melody feature database was built based on sentence, and the pitch extracted with an improved auto-correlation function, melody feature consistent with database was produced via a regularizedbased approach. Then a sentence matching method adopted to retrieve. Experimental data reveal that the new process significantly improves accuracy of humming music retrieval.