基于高频加权CQT谱的音符起始点检测  

Note onset detection based on high-frequency-weighted CQT spectrum

作　　者：林静[1] 贺前华[2] 李新超[2] 杨继臣[2] 

机构地区：[1]茂名职业技术学院机电信息系,广东茂名525000 [2]华南理工大学电子与信息学院,广东广州510641

出　　处：《计算机工程与设计》2015年第9期2485-2489,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(61301300);广东省科学技术厅基金项目(S201201000982)

摘　　要：为提高音符起始点检测的准确率,提出高频加权CQT谱能量的音符起始点检测算法。对音乐信号进行CQT变换得到音阶分布矩阵后,结合不同信号CQT分布的特点滤除噪声。在采用CQT谱能量差分进行音符起始点检测时,针对CQT变换变窗长计算带来的低频时间分辨率较低、时间误差较大可能导致误判的问题,利用音乐中谐波丰富且时间分辨率较高的特点,进行高频加权计算,突出高频成分的作用,降低检测误差。选取标注的音乐数据文件进行测试,应用高频加权CQT谱进行起始点检测的结果中,F值为0.8541,效果良好。To improve the accuracy of the note onset detection, an algorithm based on high-frequency-weighted CQT spectrum energy was proposed. The CQT transform was first performed to calculate the note distribution matrix. Then denoising filtering was conducted utilizing characteristics of distribution of CQT spectrum of noise signal. To reduce the false decision caused by low temporal resolution in low frequency bands due to varying window length, the high-frequency bands were weighted in CQT spectrum to emphasize high frequency bands utilizing rich harmonics and high temporal resolution in music. The experimental results show the F-value of the proposed algorithm is 0. 8541.

关 键 词：音符起始点 常数Q变换 频域去噪 谱能量 高频加权 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]