改进的基音检测方法及在音乐检索中的应用

基音检测是音频分析和基于内容的音乐检索中的关键技术，是基于内容音乐检索中实现哼唱检索的基础。提出一种改进的自相关函数（autocorrelation function，ACF）方法进行基音检测。从对音频信号进行去噪预处理、清浊音判断及后处理等方面对ACF进行改进，使之能够生成规整的音高变化曲线。在音乐检索的实现中，提出一种有限长度的最长公共子序列（Local Longest Common String，LLCS）方法，该方法可有效解决传统方法存在的误检问题。开发实现了一个通过哼唱/歌唱进行歌曲检索的原型系统。对大量的歌曲哼唱的实验表明，提出的改进ACF算法和LLCS算法对于提高检索正确率是正确有效的。

Improved pitch detection method and its application in musical retrieval

The pitch detection is an essential technology in audio analysis and content based music retrieval,it is foundation in query by humming.An improved method of auto-correlation function（ACF） is put forward as a pitch detection method.It includes de-noising processing, surd/sonant judgment and post-processing etc.It generates trimmed pitch curve.A Local Longest Common String（LLCS） method is proposed to solve the problem of false retrieval.A music retrieval prototype system is developed. The experiment analysis shows that the improved ACF and LLCS algorithm are effective for improving the retrieval accuracy.