高精度复调乐音识别方法  

High precision polyphonic music recognition method

作　　者：王一权 任之初 邵曦[2] 黄丽亚 WANG Yiquan;REN Zhichu;SHAO Xi;HUANG Liya(Bell Honors School,Nanjing University of Posts and Telecommunications,Nanjing Jiangsu 210023,China;School of Communication and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing Jiangsu 210023,China;College of Electronics and optical engineering&College of Flexible Electronics(Future Technology),Nanjing University of Posts and Telecommunications,Nanjing Jiangsu 210023,China)

机构地区：[1]南京邮电大学贝尔英才学院,南京210023 [2]南京邮电大学通信与信息工程学院,南京210023 [3]南京邮电大学电子与光学工程学院、柔性电子(未来技术)学院,南京210023

出　　处：《计算机应用》2023年第S02期244-249,共6页journal of Computer Applications

基　　金：江苏省高等学校大学生创新创业训练计划项目(CXXZD2022151)。

摘　　要：为解决复调乐音频识别分辨率偏低的问题,提出一种基于时频谱的复调乐音识别方法,提取复调乐音主旋律和伴奏,算法音高分辨率远超传统方法的半音阶识别。首先,用短时傅里叶变换(STFT)获得音乐信号时频谱;其次,提出一种自适应边缘失真处理方法,对时频谱进行二值化并降低音符边缘失真;再次,对二值谱进行预定位,应用改进模拟退火(SA)算法将离散域上各类变换转化至连续域,实现音符的精准定位;最后,通过基于密度的聚类算法(DBSCAN)和基频提取方法得到乐音信息。实验结果表明,对于复调音乐片段,所提算法同时具有时频高分辨率,频域平均误差小于1 Hz,时域平均误差小于50 ms。In order to boost the resolution in polyphonic music recognition,a time-frequency spectrum-based polyphonic music recognition method was proposed to extract main melody and accompaniment.The frequency resolution of the new method exceeded traditional chromatic scale identification.Firstly,Short-Time Fourier Transform(STFT)was applied to obtain time-frequency spectrum of the music signal.Then,an adaptive edge distortion method was proposed to binarize the time-frequency spectrum and reduce note edge distortion.The binaray spectrum was pre-located and improved Simulated Annealing(SA)algorithm was applied to transfer the various transformations from discrete domain to continuous doman,thus accuratly extracting fundamental frequency and harmonics.Finally,a density-based clustering algorithm and a fundamental frequency extraction method were applied to derive the musical note information.The experimental results show that the proposed method has high resolution in both time and frequency domains:the average error in the frequency domain is less than 1 Hz and the average error in the time domain is less than 50 ms for polyphonic music fragments.

关 键 词：复调音乐 时频分析 乐音识别 短时傅里叶变换 模拟退火算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]