A Time Delay Convolutional Neural Network for Acoustic Scene Classification

In recent years, demands for more natural interaction between human and machine through speech have been increasing. In order to accomplish this mission, it becomes more significant for machine to understand the human's contextual status. This paper proposes a novel neural network framework that can be applied to commercial smart devices with microphones to recognize acoustic contextual information. Our approach takes into consideration the fact that an acoustic signal has more local connectivity on the time axis than the frequency axis. Experimental results show that the proposed method outperforms two conventional approaches, which are Gaussian Mixture Models (GMMs) and Multi-Layer Perceptron (MLP), by 8.6% and 7.8% respectively in overall accuracy.