Deep CNN Framework for Environmental Sound Classification using Weighting Filters

Deep convolutional neural networks have been used to classify environmental sound recently. The classification system with high performance often requires a large well-labled dataset. The cost of tagging audio segments correctly and completely is quite high thus the deep learning models need to have high generalization ability if a weakly-tagged dataset is used. An algorithm named Weighting Filters algorithm(WF) which can be considered as an improved algorithm based on Dropout is proposed in this paper to enhance the generalization ability of models. To implement the Weighting Filters algorithm, an extra layer trained by backpropagation algorithm is introduced to produce a series of weighted filters. The simulation results show that the Weighting Filters algorithm is an effective way to improve the generalization ability of the model. Further more, a deep convolutional neural network using weighting filters algorithm is proposed for the applications of environmental sound classification. The main contributions of this paper are as follows: First, we proposed an effective algorithm WF based on Dropout, and secondly, we proposed a CNN-based framework using WF(CNN-WF) for environmental sound classification. The results obtained on ESC-50 demonstrate that the CNN-based framework we proposed has considerable performance for environmental sound classification.