D-2-DenseNet噪音鲁棒的城市音频分类模型

Noise Robust Urban Audio Classification Based on 2-Order Dense Convolutional Network Using Dual Features

为了提高噪音环境下城市音频分类系统的鲁棒性,提出了一种双特征2阶密集卷积神经网络(D-2-DenseNet)噪音鲁棒的城市音频分类模型.首先介绍了噪音添加和噪音鲁棒处理,阐述了一种双特征互补偿的算法;然后结合2阶密集卷积神经网络与自适应机制提出了一种噪音鲁棒音频分类模型:双特征2阶密集卷积神经网络.模型采用双特征互补偿自适应算法,可在特征提取与模型训练中更有针对性地提取有效音频信息,降低噪音干扰,以提高噪音鲁棒性.最后,基于Dcase2016数据集开展噪音环境下城市音频分类测试.实验结果表明,模型分类准确率分别可达77.12%、75.52%,与基线模型相比,平均分类准确率分别提高了8.51%和10.38%,验证了模型良好的噪音鲁棒性.

A noise robust urban sound event classification model based on 2-order dense convolutional network using dual features (D-2-DenseNet) is proposed,which aims at the problems of insufficient robustness of current models. Firstly,the brief introduction of the method of noise adding and robust processing is presented. Moreover,a dual feature mutual compensation algorithm and 2-order dense convolutional network is presented. Meanwhile,a noise robust urban sound event classification model based on 2-DenseNet using dual features,i. e. D-2-DenseNet is proposed. Theoretically,D-2-DenseNet combines the advantages of feature compensation and 2-order dense convolutional neural network. The dual feature mutual compensation adaptive algorithm can effectively extract audio information and reduce noise interference to improve noise robustness. Finally,in order to validate advantages of the D-2-DenseNet,this new model is exploited in the urban sound event classification based on Dcase2016 datasets. Under conditions of channel noise and environmental noise,the experiment shows that the accuracy of the network is respectively 77. 12% and 75. 52%,which has added 8. 51% and 10. 38% compared with baseline. The noise robustness of D-2-DenseNet are also effectively verified.