On the Use of Time–Frequency Reassignment and SVM-Based Classifier for Audio Surveillance Applications

Abstract

In this paper, we propose a robust environmental sound spectrogram classification approach. Its purpose is surveillance and security applications based on the reassignment method and log-Gabor filters.Besides, the reassignment method is applied to the spectrogram to improve the readability of the timefrequency representation, and to assure a better localization of the signal components. Our approach includes three methods. In the first two methods, the reassigned spectrograms are passed through appropriate log-Gabor filter banks and the outputs are averaged and underwent an optimal feature selection procedure based on a mutual information criterion. The third method uses the same steps but applied only to three patches extracted from each reassigned spectrogram. The proposed approach is tested on a large database consists of 1000 sounds belonging to ten classes. The recognition is based on Multiclass Support Vector Machines.

A mixture model-based real-time audio sources classification method

“…The second one uses machine learning algorithm to model the sound classes based on the previous features. Commonly used algorithms are Gaussian Mixture Models (GMM) [6,7], Support Vector Machines (SVM) [8,9], Hidden Markov Models (HMM) [10,11] or Neural Networks (NN) [12,13]. More recently, research focuses on neural networks and uses them for features extraction and classification at the same time.…”