基于倒谱分析的弦乐和打击乐的源分离  

Harmonic Percussive Sound Source Separation Based on Cepstrum Analysis

作　　者：吴礼福[1,2] 孙芯年 WU Lifu;SUN Xinnian(School of Electronic&Information Engineering,Nanjing University of Information Science&Technology,Nanjing 210044;Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,Nanjing 210044)

机构地区：[1]南京信息工程大学电子与信息工程学院,南京210044 [2]江苏省大气环境与装备技术协同创新中心,南京210044

出　　处：《计算机与数字工程》2024年第8期2524-2529,共6页Computer & Digital Engineering

摘　　要：针对音乐中弦乐和打击乐的分离(Harmonic Percussive Sound Separation,HPSS)问题,基于弦乐在时域上跨度长、频域上分布窄,而打击乐在时域上跨度短、频域上分布宽的特点,论文研究了一种倒谱滤波加后处理的分离方法。该方法在倒谱域设计滤波器将弦乐和打击乐初步分离后,对打击乐中残留的弦乐部分进行后处理,最后再变换回时域信号。采用音频信号客观评价(Perceptual Evaluation of Audio Quality,PEAQ)算法对分离后的音乐进行评估,结果表明该方法能有效地分离出打击乐和弦乐,同时无需数据驱动类分离方法中大量训练样本的支撑。Aiming at the problem of harmonic percussive sound separation,this paper introduces a cepstrum filtering and post-processing separation method based on the characteristics that harmonic has long span in time domain and narrow distribution in frequency domain,while percussion has short span in time domain and wide distribution in frequency domain.In this method,the harmonic and percussion is preliminary separated by a filter designed in cepstrum domain,then the residual harmonic part in percussion is post processed,and finally it transforms back to time domain signal.Perceptual evaluation of audio quality is used to evaluate the separated music,the results show that this method can effectively sort out harmonic and percussion without the support of a large number of training samples in the data-driven class separation method.

关 键 词：弦乐 打击乐 音乐分离 倒谱 

分 类 号：TB53[理学—物理]