[1]何 丽,袁 斌.利用长短期记忆网络进行音乐流派的分类[J].计算机技术与发展,2019,29(11):190-194.[doi:10. 3969 / j. issn. 1673-629X. 2019. 11. 038]
　HE Li,YUAN Bin.Classification of Music Genres Using Long and Short-term Memory Network[J].,2019,29(11):190-194.[doi:10. 3969 / j. issn. 1673-629X. 2019. 11. 038]
点击复制

利用长短期记忆网络进行音乐流派的分类()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
29
期数:
2019年11期
页码:
190-194
栏目:
应用开发研究
出版日期:
2019-11-10

文章信息/Info

Title:
Classification of Music Genres Using Long and Short-term Memory Network
文章编号:
1673-629X(2019)11-0190-05
作者:
何 丽; 袁 斌
北方工业大学 计算机学院,北京 100144
Author(s):
HE Li; YUAN Bin
School of Computer,North China University of Technology,Beijing 100144,China
关键词:
音乐分类; 长短时记忆网络; 梅尔倒谱系数; 频谱质心; 频谱对比度
Keywords:
music classification; long short-term memory network; Mel frequency cepstral coefficient; spectral centroid; spectral contrast
分类号:
TP399
DOI:
10. 3969 / j. issn. 1673-629X. 2019. 11. 038
摘要:
针对传统的基于人工标注的文本音乐分类存在人工标注成本高,易于出错,没有涉及到音乐本身的内容的问题,提出了一种基于音乐内容的分类方法,将深度学习中的长短时记忆网络(LSTM)应用到音乐流派分类中。 从包含 10 种音乐流派的 1 000 首歌曲中提取出梅尔倒谱系数,频谱质心和频谱对比度三个特征,将提取出来的特征数据输入到 LSTM 网络中进行训练,输出每种音乐类别的概率。 对此,进行了三次实验。 第一次是将梅尔倒谱系数,频谱质心作为特征数据输入到 LSTM 网络中,第二次是以频谱对比度和频谱质心作为特征数据,第三次是将梅尔倒谱系数,频谱质心和频谱对比度作为特征数据。 从实验结果上看,当梅尔倒谱系数,频谱质心和频谱对比度作为特征数据时,模型的分类效果最好,分类准确率最高。 实验结果表明,该方法在准确率上比玻尔兹曼机和卷积神经网络等方法都有所提升。
Abstract:
In view of the problems of traditional music classification of text based on manual labeling,such as high manual labeling cost,error-prone and not involving music content,we propose a classification method based on music content,which applies long and short-term memory network (LSTM) in deep learning to music genre classification. Three features including Mel cepstrum coefficient,spectral centroid and spectral contrast are extracted from 1 000 songs of 10 music genres. The extracted feature data is input into LSTM network for training,and the probability of each music category is output. Three experiments are carried out on this. The first is to input the Mel frequency cepstral coefficient and spectral centroid as feature data into the LSTM network. The second is to use spectral contrast and spectral centroid as feature data,and the third to use Mel frequency cepstral coefficient,spectral centroids,and spectral contrast are as feature data. From the experimental results,when Mel frequency cepstral coefficient,spectral centroids and spectral contrast are used as feature data,the model has the best classification results,the highest classification accuracy. Experiment shows that the method proposed is more accurate than Boltzmann and convolutional neural networks.

相似文献/References:

[1]李东欣,禹 龙,田生伟,等.注意力机制的 LSTM-DBN 维语人称代词指代消解[J].计算机技术与发展,2019,29(07):33.[doi:10. 3969 / j. issn. 1673-629X. 2019. 07. 007]
　LI Dong-xin,YU Long,TIAN Sheng-wei,et al.Attention Mechanism of LSTM-DBN Uyghur Personal Pronoun Anaphora Resolution[J].,2019,29(11):33.[doi:10. 3969 / j. issn. 1673-629X. 2019. 07. 007]
[2]王振业,叶成绪*,王文韬,等.基于 LSTM-Att 方法的音乐流行趋势预测[J].计算机技术与发展,2020,30(09):188.[doi:10. 3969 / j. issn. 1673-629X. 2020. 09. 034]
　WANG Zhen-ye,YE Cheng-xu*,WANG Wen-tao,et al.Music Trend Forecast Based on LSTM-Att Method[J].,2020,30(11):188.[doi:10. 3969 / j. issn. 1673-629X. 2020. 09. 034]
[3]吕 鹏,曹江涛,姬晓飞*.基于多任务学习的火灾检测算法[J].计算机技术与发展,2023,33(01):56.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 009]
　LYU Peng,CAO Jiang-tao,JI Xiao-fei*.Fire Detection Algorithm Based on Multi-task Learning[J].,2023,33(11):56.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 009]
[4]王 健,宋 颖,吴 涛.基于 LSTM 网络与误差补偿的预测模型[J].计算机技术与发展,2023,33(03):133.[doi:10. 3969 / j. issn. 1673-629X. 2023. 03. 020]
　WANG Jian,SONG Ying,WU Tao.Prediction Model Based on LSTM Network and Error Compensation[J].,2023,33(11):133.[doi:10. 3969 / j. issn. 1673-629X. 2023. 03. 020]
[5]陈 雄,王海晨.基于 ISSA-LSTM 模型的短时交通流预测[J].计算机技术与发展,2023,33(04):198.[doi:10. 3969 / j. issn. 1673-629X. 2023. 04. 029]
　CHEN Xiong,WANG Hai-chen.Research on Traffic Flow Prediction Based on ISSA-LSTM Model[J].,2023,33(11):198.[doi:10. 3969 / j. issn. 1673-629X. 2023. 04. 029]

更新日期/Last Update: 2019-11-10