Environment Sensing using Acoustic Data

Abstract

With the recent rise of deep learning approaches for artificial intelligence, we came to realize that developing computer systems that are reliably capable of understanding visual scenes and recognizing speech is finally possible. Unfortunately, little attention has been given to the idea of developing computer systems that can emulate the way we use our sense of hearing to make sense of what’s around us. Making intelligent systems that can reliably recognize environmental sounds will cause a paradigm shift in technological areas such as audio surveillance, noise pollution analysis, audio-based search engines, and hearing aids. This work digs into the use of convolutional neural networks (CNN) for robust environmental sounds recognition, while also proposing a method for dealing with small datasets called stochastic continuous data augmentation. Obtained results have been compared with some previous related works