A deep convolutionary network for automatic detection of audio events

In the last years, a growing interest of the scientific community has been devoted to the design of efficient and effective solutions for automatically identifying sounds of interest in the field of audio surveillance. In this paper, we propose an event detection architecture based on a convolutional neural network. The network takes in input directly the raw sampled waveform, in order to reduce the information loss introduced by any type of preprocessing and is able to automatically learn the most important frequencies for the sounds of interests to be recognized. We evaluate the performance of the proposed approach over the MIVIA Audio Events dataset, widely adopted for benchmarking purposes, achieving the best results and, therefore, confirming its effectiveness with respect to state of the art methodologies, based on both traditional machine learning and deep learning methodologies.