基于机器学习的音频分类  

Audio classification based on machine learning

作　　者：熊华煜 余勤[1] 任品 雒瑞森 XIONG Hua-yu;YU Qin;REN Pin;LUO Rui-sen(College of Electrical Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学电气工程学院,四川成都610065

出　　处：《计算机工程与设计》2021年第1期156-160,共5页Computer Engineering and Design

基　　金：校企合作基金项目(17H1199、19H0355);中国博士后科学基金面上基金项目(2017M612958)。

摘　　要：为施行有效的音频分类以高效率处理日渐复杂的音频信息,研究采用包含多种神经网络在内的5种机器学习模型,实现多种决策下的音频分类以寻找最优模型,基于分类准确度对各模型分类效果进行评估,在使用正则化方法保证模型泛化能力的条件下,通过比较和实验,挖掘并验证出了相对最优的模型——卷积神经网络音频分类模型及对应参数,为现有音频分类模型的进一步优化提供了参考方向。To implement effective audio classification for efficient audio information process which is becoming increasingly complex,five kinds of machine learning models including neural network were built respectively to find the best classification model,and the accuracy of classification of each model was evaluated.The relatively better model,convolutional neural network for audio classification,and its corresponding best parameters were recommended and verified based on experiment and comparison under dropout regulation to guarantee the ability of generalization.The new reference orientation was provided to optimize exis-ting classification model.

关 键 词：多媒体技术 机器学习 音频分类 神经网络 正则化 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]