查看文章

This paper is focused on the surveillance system based on the acoustic events detection. Sounds representing potential danger are investigated. Detection of two types of acoustic events (gun shot and breaking glass) was done. In this work, different intensity of background sounds were evaluated. For representing of acoustic events MFCC, Audio Spectral Centroid, Audio Spectral Spread and Audio Spectral Flatness were used. Hidden Markov Models (HMMs) were used to classify the input patterns. The system performance was evaluated by miscellaneous intensity of background for testing recording.