“A Multilingual Audio-based Threat Detection System”

Dr. Rekha B Venkatapur1, N.Kiran Raj2

1 Professor and Head, Department of Computer Science and Engineering, KSIT, India

2 Student, Department of Computer Science and Engineering, KSIT, India

Abstract:
Manual safety tools like SOS buttons and emergency calls often do not work during emergencies because people may be too scared physically trapped or unable to act. Recent advancements in learning have enabled the creation of smart voice-based systems that can detect distress, aggression and unusual sounds in human speech. This project introduces a hands- mobile app that supports English, Hindi and Kannada and continuously listens to surrounding audio to detect distress situations. The app uses features to send emergency alerts along with the users GPS location. A dataset of 200 clips in three Indian languages was prepared to reflect real-world situations, including screams, crying and gunshots. The app uses learning models and the results show that it can reliably identify threats across languages and environments. The system provides a safety solution and aligns with goals 3, 5 and 16 of the Sustainable Development Goals. The app records audio intervals uploads them to a backend, for prediction and triggers instant alerts through SMS or notification.

Keywords:

Safety, Privacy, Security, Detection, Artificial Intelligence, Emergency, SOS, Alert system, CNN-LSTM, Threat detection.