基于最大最小距离法的节拍跟踪

Beat Tracking Based on Max-Min Distance Means

将端点检测与聚类算法结合,提出一种基于最大最小距离法的节拍跟踪算法.首先,将音乐信号分解成多个频率互不重叠的子带进行频谱分析,分别利用半波整流,求和得到最终的端点强度曲线.其后,根据能量谱找到起始节拍点.最后,根据最大最小距离法并利用音乐速度与节拍的关系,对端点强度曲线峰值进行有效聚类,识别出节拍点.实验结果表明,该算法识别节拍点准确有效,4评估指标P-score、Cemgil、CMLc和AMLt分别达到57.355,10、38.705,37、17.152 40和47.259 12,与其他算法相比综合性能较好.

In this paper,combining onset detection and clustering algorithm,a novelbeat tracking algorithm based on max-min distance(MMD) means was proposed. First,the spectrum of music signal was decomposed into several non-overlapping sub-bands. Second,by utilizing the half-wave rectifier on these sub-bands respectively,the final onset strength curve was found. Then,the first beat was discovered based on the energy spectrum of the starting point. Finally,based on MMDmeans,the beats of the music signal were identified according to the relationship between music tempo and beat,together with effective clustering of curve peaks of onset strength. Experimental re-sults proved that the proposed algorithm can track beats accurately. The four evaluation indicators of the algorithm P-score,Cemgil,CMLc and AMLt,reached 57.355,10,38.705,37,17.152,40 and 47.259,12,respectively. Com-pared with other algorithms,the proposed algorithm possesses better comprehensive performance.