DeepESC网络的环境声分类方法研究

为进一步提升环境声分类的识别率，提出了一种仿深度隐藏身份特征（Deep Hidden Identity Feature，DeepID）网络连接方式的卷积神经网络——深度环境声分类网络（Deep Environment Sound Classification，DeepESC）。DeepESC网络共有六层——三层卷积层、两层全连层以及一层聚合层，为使网络在自动抽取高层次特征的同时能有效地兼顾低层次特征，网络将三层卷积层的输出聚合为一层，该层充分包含不同层次的特征，提升了卷积神经网络的特征表达能力。ESC-10和ESC-50数据集上的仿真结果表明：在相同的识别框架下，与随机森林分类器相比，本文网络识别率分别平均提升了7.6%和22.4%，与传统的卷积神经网络相比，识别率分别平均提升4%和2%，仿真实验验证了本文分类器的有效性。

Environmental sound classification using DeepESC convolutional neural networks

To improve the accuracy of environmental sound classification, a new convolutional neural network named DeepESC, which imitates the connection of DeepID network, is proposed. DeepESC is composed of three convolution layers, two fully connected layers and one concatenate layer. To extract both high-level features and low-level features effectively, a concatenate layer is designed to join all convolution layers'' output together, which comprises all features of different levels in the DeepESC network. Experimental results on ESC-10 and ESC-50 data sets show that, compared with random forest classification in same conditions, the accuracy of DeepESC is improved by 7.6% and 22.4% respectively, and by 4% and 2% respectively compared with the traditional convolutional neural network.