Abstract
Video analysis has become a standard feature of many security cameras. However, built-in audio analytics continues to be quite rare despite the presence of both the audio channel itself in the devices and the available computing power for processing audio data. Audio analytics has some advantages over video analytics such as cheaper devices and maintenance costs. Furthermore, when the system is running in real-time, the audio data stream is significantly smaller in volume than the data stream from video cameras and makes it more loyal requirements for the bandwidth of the data channel. Audio analytics systems can be particularly useful for urban surveillance with the start of automated broadcasting live video to the police console from the scene of an explosion and shooting. Audio analytics technologies can also be used to study video recordings and determine events. This article proposes a method for automatic detection of pulse sounds that signifies critical situation in audio signals based on Support Vector Machine learning models. The models were able to classify sounds from events such as gunshot, broken glass, explosion, siren, cry and dog barking with accuracy ranges from 95% to 81 %.