基于ResNet网络的音乐类型分类及标签标记模型研究  

Research on music type classification and label labeling model based on ResNet network

作　　者：易伶 YI Ling(School of Art,Shangluo University,Shangluo 726000,Shaanxi Province,China)

机构地区：[1]商洛学院艺术学院,陕西商洛726000

出　　处：《信息技术》2025年第2期80-85,91,共7页Information Technology

摘　　要：个性化的音乐推荐服务能为人们带来情感上的寄托和提供更多的便捷。然而,音乐类型的分类和标签的标注需要挖掘大量的音乐特征,耗费时间和人工成本。基于此,该研究将音频数据经过傅里叶变换转换为Mel频谱图,将残差胶囊网络与门控循环单元相结合,构建音乐类型分类模型。同时,基于残差胶囊网络,引入视觉注意力机制,构建音乐标签标记模型。结果显示,音乐类型分类模型在GTZAN和FMA-smile数据集上的分类准确率能达到89.8%和85.2%。该研究结果能为音乐推荐平台的个性化服务提供一定的参考。Personalized music recommendation services can bring emotional support and provide more convenience to people.However,the classification of music types and labeling require mining a large number of music features,which consumes time and labor costs.Based on this,this study converts audio data into Mel spectrograms through Fourier transform,and combines residual capsule networks with gated loop units to construct a music type classification model.Meanwhile,based on the residual capsule network,a visual attention mechanism is introduced to construct a music label labeling model.The results show that the classification accuracy of the music type classification model on the GTZAN and FMA-smile datasets can reach 89.8%and 85.2%,respectively.The results of this study can provide some reference for personalized services of music recommendation platforms.

关 键 词：音乐分类 傅里叶变换 残差胶囊网络 门控循环单元 视觉注意力机制 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]