Automatic threat classification using multiclass SVM from audio signals

Analysis of sound signals in terms of threat detection, diagnosis and classification is an important part of modern surveillance systems. The subject of this study is the analysis, detection and classification of threat sounds representing events widely considered to be dangerous. Currently available solutions have not been created with threat sounds in mind and are mostly speech-recognition solutions. The proposed algorithm is based on mel-cepstral coefficients and SVM, which is commonly used in identifying fragments of images and other patern recognition issues. The decision-making system presented here uses this method for audio analysis, and will also create a new area for applications.