摘要
针对单一特征建立的音乐流派分类模型导致误判、遗漏、错分的不足及其处理速度慢、效率低等问题,提出了一种基于卷积神经网络的音乐流派分类方法。该算法首先采用倒谱系数提取音频的MFCC特征矩阵,以其特征值作为CNN神经网络的输入量对音频信号进行训练,获取最优分类器用以作为训练器。将经典、乡村、重金属和摇滚四种音乐流派的音频信息通过最优分类器进行仿真实验。实验结果表明:卷积神经网络分类的平均分类效率可达88%,处理速度明显提升,降低误分类及错判率。
Aiming at the problems of misjudgment,omission and misclassification caused by the music genre classification model based on single feature,and its slow processing speed and low efficiency,this paper proposes a music genre classification method based on convolution neural network.Firstly,the cepstrum coefficient extracts the MFCC feature matrix of the audio,and the eigenvalue is used as the input of the CNN neural network to train the audio signal,and the optimal classifier is obtained as a training device.The audio information of the four music genres of classic,country,heavy metal and rock is simulated by the optimal classifier.The experimental results show that the average classification efficiency of convolutional neural network can reach 88%,the processing speed is obviously improved,and the misclassification and misjudgment rate are reduced.
作者
陆欢
Lu Huan(University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《电子测量技术》
2019年第21期149-152,共4页
Electronic Measurement Technology
关键词
音乐分类
音频特征
特征提取
CNN神经网络
music classification
feature extraction
convolutional neural network
