HYBRID LSTM–BIGRU MODEL FOR HIGH-ACCURACY ENVIRONMENTAL THREAT DETECTION WITH INTELLIGENT ALERT ROUTING

Abstract

This follow-up study enhances the first LSTM architecture for real-time audio-based threat detection by adding a Bidirectional Gated Recurrent Unit (BiGRU) layer. To improve the model's comprehension of complicated sound patterns like gunshots, cries, and glass breaking, the BidirectionalGRU records both forward and backward temporal dependencies in ambient audio. This leads to improved accuracy and resilience in the system when contrasted with earlier uses of standalone LSTM and CNN models. Improving the threat alert mechanism's practical applicability, we've made it possible to dynamically change the recipient email address. This way, in an emergency, notifications will be delivered to the most relevant contact right away. Also included is a web app built using Flask that allows administrators to easily submit audio files and see the results of threat categorization instantly. In addition to enhancing the detection model's intelligence, this expansion makes the safety system far more accessible, flexible, and useful in the real world