基于VGGish网络的音频信息情感智能识别算法  

Intelligent recognition algorithm of audio information emotion based on VGGish network

作　　者：张志超[1] 李晓燕 ZHANG Zhichao;LI Xiaoyan(Luxun School of the Arts,Yan’an University,Yan’an 716000,China;School of Mathematics and Computer Science,Yan’an University,Yan’an 716000,China)

机构地区：[1]延安大学鲁迅艺术学院,陕西延安716000 [2]延安大学数学与计算机科学学院,陕西延安716000

出　　处：《电子设计工程》2022年第4期26-30,共5页Electronic Design Engineering

基　　金：2020年度陕西省教育厅科学研究计划项目(20JK0982)。

摘　　要：针对目前对于音频信息情感的研究较少以及对音频情感进行智能分析识别缺乏成熟技术的问题,文中提出了一种基于VGGish网络的音频信息情感智能识别算法。该算法通过VGGish网络提取音频信息的特征,利用降维可视化的方法不断调整不同情感音乐数据集分布,通过SVM和LSTM模型实现了音乐情感分类。算法测试结果表明,两种分类模型均能够对音频信息情感进行准确有效地分类,尤其是LSTM模型对于音频信息情感分类的平均准确率可达90.12%。In view of the lack of research on audio information emotion and the lack of mature technology for intelligent analysis and recognition of audio emotion,an intelligent recognition algorithm of audio information emotion based on VGGish network is proposed.The algorithm extracts the features of audio information through VGGish network,adjusts the distribution of different emotional music data sets by using the method of dimension reduction and visualization,and realizes the music emotion classification through SVM and LSTM model.The algorithm test results show that the two classification models can classify audio information emotion accurately and effectively.Especially the LSTM model,the average accuracy of audio information emotion classification can reach 90.12%.

关 键 词：VGGish网络 卷积核 数据标注 数据转换 可视化 音频信息 

分 类 号：TN971[电子电信—信号与信息处理] TP391[电子电信—信息与通信工程]