【摘要】 卷积神经网络(CNN)辨别频域-时域模式的能力使其适合于环境声音分类。然而数据的相对稀缺使该方法的应用受限。所以使用数据增强与卷积神经网络结合的方法来克服这一难点。首先,提出使用音频数据增强来增加训练数据,然后提出了一种卷积神经网络模型进行分类。所提出的方法对于环境声音分类的准确率达到了79.5%,这种方法既优于没有增强的CNN模型也优于具有增强的SVM模型。更多还原

【Abstract】 The ability of the Convolutional Neural Network to discern the frequency-time domain mode makes it suitable for environmental sound classification. However, the relative scarcity of data limits the application of this method. Data enhancement combined with convolutional neural networks is used to overcome this difficulty. First, it is proposed to use audio data enhancement to increase training data. Then, a convolutional neural network model is proposed for classification. The accuracy of the method proposed for the classification of environmental sounds reaches 79.5%. This method is better than the general CNN model and the enhanced SVM model.更多还原