Paper Title
Exploiting Autocorrelation Feature in Deep Learning-Based Audio Classification

This paper concerns a feature parameter for audio classification based on deep neural network (DNN) which can be applied for environment classification in internet of things (IoT) services. In image classification, the raw pixel data is successfully used as the input data of DNN. In audio classification, however, the raw audio signal is not suitable for input feature parameter of DNN because it has complex characteristics such as temporal variation and phase. We propose a new approach of using autocorrelation coefficients as input feature parameters instead of raw audio signals. The experimental results show that the proposed feature gives improved performance compared with the raw audio signal in DNN-based audio classification. Index terms - Autocorrelation, Audio Classification, Deep Neural Network, Internet of Things (IoT).