DETECTION OF EMERGENCY SITUATION USING DEEP LEARNING SOUND RECOGNITION METHODS

DOI: https://doie.org/10.10399/JBSE.2025141993

Avijit Bose, Bidisa Chowdhury, Rajdeep Roy, Ishika Ghosh, Fazal Hussain, Ahana Mukherjee, Satyajit Chakrabarti

Keywords:

Deep learning (DL), Anomalous event detection, Household safety, Emergency detection, Audio data monitoring, Convolutional Neural Networks (CNNs), Mel Spectrogram, Audio pattern analysis, Indoor environment monitoring, Hazardous sound classification

Abstract:

Machine learning (ML) and deep learning (DL) technologies enable sound recognition systems to perform their functions.The detection of anomalous events across different environments depends critically on sound recognition technology which is especially important for security and household safety. Emergency detection within home environments demands special attention for elderly residents.The safety of elderly individuals and children remains a crucial issue as they may struggle to communicate in emergencies. Automated systems capable of real-time detection that recognize distress signals and accidents enable effective monitoring of household audio data.The proposed system monitors household audio to identify life-threatening situations through detection of incidents such as screaming or shattering noises or gunshots. Advancements in deep learning, particularly  CNNs have created the capability to achieve highly precise recognition of sound patterns. By examining audio patterns researchers can differentiate between everyday household sounds and signals that indicate potential danger. Our research introduces a deep learning model which tackles the problem of recognizing emergency sounds within indoor spaces. The research offers a sound recognition model based on deep learning that monitors occupant safety and detects potential emergencies. Experiments are conducted using audio data originated from both real-world environments and online[18] sources. Our model achieved an accuracy of 90%, with promising precision and recall rates. The[27] proposed system shows potential to improve safety by generating timely alerts establishes a foundation for future research on sound recognition integration.

PDF