基于卷积循环神经网络的中国民族复音音乐的乐器活动检测  

Instrument Activity Detection of China National Polyphonic Music Based on Convolutional Recurrent Neural Network

作　　者：李子晋 蒋超亚 陈晓鸥[2] 马英浩 韩宝强 LI Zijin;JIANG Chaoya;CHEN Xiao'ou;MA Yinghao;HAN Baoqiang(Musicology Department,China Conservatory of Music,Beijing 100101,China;Wangxuan Institute of Computer Technology,Peking University,Beijing 100080,China)

机构地区：[1]中国音乐学院音乐学系,北京100101 [2]北京大学王选计算机研究所,北京100080

出　　处：《复旦学报（自然科学版）》2020年第5期511-516,共6页Journal of Fudan University：Natural Science

基　　金：国家艺术基金(01020120180529564031)。

摘　　要：针对中国民族复音音乐的乐器活动检测问题,提出了一种基于卷积循环神经网络(CRNN)的复音乐器活动检测方法,该方法属于事件检测类,在秒级的时间分辨率上识别乐器活跃的起止时间及乐器种类.同时,在中国音乐学院的DCMI数据库基础上,构建了3种不同的面向10种中国民族乐器的复音数据集进行训练和评估.通过实验,我们将CRNN模型与CNN模型进行了比较,验证了模型的特点和优势.Aiming at the instrument activity detection of Chinese national polyphonic music,apolyphonic musical instrument activity detection method based on Convolutional Recurrent Neural Network(CRNN)is proposed.This method is a kind of event detection method which identifies the starting and ending time of musical instruments and musical instrument types in the second temporal resolution.At the same time,based on the Database of China Conservatory of Music(DCMI),three different polyphonic instrument detection datasets which contain 10 kinds of China national instruments were constructed for training and evaluation.Through experiments,we compare the CRNN model with the CNN model,and verify the advantages of the CRNN model.

关 键 词：卷积循环神经网络 中国民族音乐 乐器活动检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]