High performance neural network based acoustic scene classification

Abstract

Acoustic Scene Classification (ASC) has become an integral component in applications such as smart hearing aids, user alert applications for physically challenged persons and robot based navigation. The Deep Learning (DL) techniques such as Deep Neural Network (DNN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) etc. improve the accuracy and efficiency of ASC but increase computational complexity. This paper has explored the possibility of using neural network as classifiers for ASC. The observations show that an appropriately trained simple neural network can achieve similar performance as DL techniques. The developed model utilizes Mel Frequency Cepstral Coefficients (MFCC) for feature extraction and has been verified on TUT Acoustic Scenes 2016 dataset. The developed model has attained 14.3% better accuracy than existing DL models for frame based analysis.