Abstract:

To reduce the limited range when used to multiple kinds of music signals and high computing complexity of existing note onset automatic detection methods, a note onset detection algorithm based on differential all phase Mel Frequency Cesptrum Coefficients(MFCC) is presented. All phase preprocessing alleviates spectrum ambiguity because of spectrum leakage. And differential Mel frequency cesptrum considered nonlinearity of human ear when responding to different frequencies and dynamic property of music signal. Experimental result demonstrates this algorithm is fit for multiple kinds of music and has perfect general detection performance with lower computing complexity compared with those of High Frequency Content(HFC) and Independent Component Analysis(ICA).

Key words: note onset detection, Mel Frequency Cesptrum Coefficients(MFCC), all phase preprocessing, Music Information Retrieval(MIR)

摘要：

针对现有的音符起点自动检测方法难以适用于多类音乐信号，计算复杂度较高等问题，提出一种基于差分全相位MFCC的检测算法。通过全相位预处理减小频谱泄露引起的频谱模糊，差分Mel频率倒谱考虑人耳对音乐不同频率响应的非线性特性和音乐信号的动态音乐特征。实验结果表明，与公认综合检测效果好的HFC和ICA等方法相比，该方法计算复杂度小，适用音乐信号类型广，具有更优的综合检测性能。

关键词: 音符起点检测, Mel频率倒谱系数, 全相位预处理, 音乐信息检索

CLC Number: 

• TP18