一种适用于非特定哼唱方式的起音点检测算法

摘要

A pitch based Spectral Flux(SF) algorithm is proposed for detecting the onset times of musical notes in humming signals. By using sliding window average filtering, SF algorithm and the feature of pitch change, the proposed algorithm can effectively filter the interference of redundant spectral energy, greatly reduce segmentation errors and achieve good detection improvement for those consecutively singing words. Experimental results show that the detection accuracy rate reaches 80%, better than other well used detection algorithms.%针对现有音符起音点检测算法对非特定哼唱方式分割效果不佳的现状,提出一种新的基于音高的频谱差异算法.结合哼唱音高的变化特性,利用频谱差异算法、滑动窗平均滤波滤除冗余频谱能量干扰,降低过分割、误分割的检测错误.实验结果表明,该算法的检测准确率达80％,优于现有起音点检测算法.