A Voting-Based Technique for Acoustic Event-Specific Detection

Several works attempted to address the problem. In general, these employ simple frame-level presentations and a variety of classification algorithms. Typically, individual events are modelled as Hidden Markov Models (HMM), and a speech recognition framework is employed to detect them [4]. The audio segments can also be characterized by the Gaussian population histograms derived from a Gaussian Mixture Model (GMM), and the detection is performed as classification task using GMMs [5]. In another work, Support Vector Machines (SVM) are directly used over feature vectors derived from audio signals [2].