ML-based detection and differentiation of gunshot noises on mobile devices for smart environments

In the current era, smartphones are utilized for recording purposes on a continuous basis. Videos are continually being captured, particularly at events, and brief clips are being shared on social media platforms such as Instagram and TikTok. At events and on special occasions such as New Year’s Eve, as well as during hunting season in forest areas, there are instances when individuals unintentionally record audio or video footage without being able to identify the source. In Europe, where there are stringent regulations governing the use of firearms, it is important to distinguish between shots fired from weapons utilized by security agencies, hunting weapons, and alarm weapons. Furthermore, we live in a world where smartphones and smartwatches are always with us and are capable of recognizing sounds, detecting falls, and so on. To enhance safety and enable more effective situational assessment, as well as facilitate automatic emergency alarm activation in the event of gunshots, we present an AI-based smartphone application capable of recognizing and differentiating between gunshots. The system is trained and optimized in a testbed using real recordings of shots at shooting ranges. The smartphone is used as a recording device to depict scenarios that are as practical and realistic as possible. The two-level approach allows us to optimize the solution.