Environmental Sound Classification Using Deep Convolutional Neural Networks and Data Augmentation

Abstract

This work is about environmental sound classification by deep convolutional neural networks and data augmentation. Data augmentation is applied to increase the labeled training dataset. Data augmentation process improves the performance of audio classification. In this paper, first we present a strategy for generating a deep convolutional neural network (CNN) framework for environmental sound analysis with Urban-sound8K audio dataset. Secondly we analyze the performance of data augmentation methods on Urbansound8K audio dataset and compare the performance of CNN with different data augmentation methodologies. Data augmentation is basically a deformation technique. By this approach we can increase the number of dataset elements into its multiples. Here, compare the performance of different augmentation method to identify which one is the best augmentation technique for environmental sound analysis. Different types of data augmentations were applied to the dataset in the previous works. We introduce a new data augmentation method using LPCC feature.