Abstract:
This paper describes an approach for an audio event detection system in noisy environments. The system specifically focuses on classification of an audio event as gunshot, scream or ambient noise. For discriminating gunshot from noise and scream from noise, two parallel Gaussian Mixture Model (GMM) classifiers are applied. Acoustic features such as zero-crossing rate, mel frequency cepstral coefficients, spectral flatness measures are firstly extracted to train GMM classifier. Each GMM classifier is trained using different set of audio features. To reduce the false detection rate, the decision that an event (gunshot or scream or noise) is taken by computing logical OR of the two classifiers. The efficiency of this scheme is investigated over audio recordings taken from internet repositories. The experimental results show that the overall accuracy of the system is as high as 97%.
