基于离散谐波变换的西洋乐器音色特征提取方法  

Timbre Feature Extraction of Western Musical Instrument Based on Discrete Harmonic Transform

作　　者：孙聪珊 杨婧 马琳[1] 李海峰[1] SUN Congshan;YANG Jing;MA Lin;LI Haifeng(School of Computer Science and Technology,Harbin Institute of Technology,Harbin,Heilongjiang 150001,China)

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《复旦学报（自然科学版）》2020年第5期531-539,共9页Journal of Fudan University：Natural Science

基　　金：国家自然科学基金(61671187);深圳市基础研究项目(JCYJ20180507183608379)。

摘　　要：音色是乐器识别的依据,考虑到乐音信号的谐波对音色的物理意义,本文提出具有良好稀疏性的离散谐波变换(Discrete Harmonic Transform,DHT)用于乐音信号谐波结构提取,并通过DHT构建了描述乐器音色的DHT特征.通过将本文提出的特征应用于不同复杂度的乐器音频数据集,验证了本文所提出的DHT方法的有效性,在此基础上构建的DHT特征能很好地反应乐器音色的特征,为乐音信号的特征提取提供了新思路.进一步,将本文提出的DHT特征与线性倒谱预测系数(Linear Prediction Cepstrum Coefficient,LPCC)和Mel频率倒谱系数(Mel Frequency Cepstrum Coefficient,MFCC)构成的组合特征分别应用于乐器演奏的单音和乐曲片段数据集的乐器识别中,与现有算法相比,新算法在较低特征维数下达到了更高的乐器识别的准确率.Timbre is the basis of musical instrument recognition.Considering the physical significance of harmonic of musical signal to timbre,this paper proposes Discrete Harmonic Transform(DHT)with good sparse to extract harmonic structure of musical signal,and constructs the timbre expression spectrum of musical instrument timbre through DHT.By applying the features proposed in this paper to the audio data sets of musical instruments with different complexity,the effectiveness of the DHT method proposed in this paper is proved.DHT features can reflect the characteristics of musical instrument timbre very well,which provides a new idea for feature extraction of musical signal.Furthermore,the combination of DHT features,Linear Prediction Cepstrum Coefficient(LPCC)and Mel Frequency Cepstrum Coefficient(MFCC)proposed in this paper are applied to the musical instrument recognition.Compared with the existing algorithm,it achieves a higher recognition rate under a lower feature dimension.

关 键 词：智能乐器识别 音色 谐波结构 离散谐波变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]