Trends of Sound Event Recognition in Audio Surveillance: A Recent Review and Study

Abstract Identification of environment sounds plays a key role in security and surveillance aspects. In this paper, we present a review and recent advances in a sound event recognition (SER) problem and discuss different deep learning approaches applied for SER. SER is the process of extracting the pertinent features from the given sound and classify it into various categories. This SER task has attracted a lot of attention in recent times such as smart homes, cough sound detection, etc. Apart from other surveys, this survey focuses on SER task and carried out an extensive study of few baseline methods on Environmental Sound Classification 50 data set, which consists of 50 sound classes. From this data set, we have extracted Mel-frequency cepstral coefficient features and fed as input to classifiers such as support vector machine, artificial neural network, and convolution neural network (CNN). Based on this study and experimentation, we explore various pros and cons over this SER task. Deep learning techniques are used to improve the recognition rate of SER task. It is evident from the results reported in the DCASE challenge 2016–2017 data set that CNNs are proven to be effective for SER.