Emergency Sound Classification and Visual Alert System for Enhanced Situational Awareness

This research introduces an audio classification system designed to enhance situational awareness for individuals with hearing impairments. The system recognizes emergency sounds and presents corresponding visual alerts. Utilizing Google's pre-trained YAMNet model, it accurately identifies crucial sounds such as ambulance, firetruck, and police sirens, as well as railroad crossing and other danger alarms, distinguishing them from typical background noise. The system, deployed on an Intel Core i5 processor with an NVIDIA GeForce RTX 2050 (compute capability 8.6), extracts audio features from sound files and classifies them using a trained model. Each identified sound category triggers a specific visual indicator on a graphical interface. To enhance robustness in noisy environments, a noise reduction preprocessing step is applied, improving classification accuracy. Testing demonstrates a 93.68% accuracy rate in emergency sound detection with an average latency of 2.9 ms for sound classification. The simulation latency ranges between 120 ms and 200 ms. These results highlight the system's potential for real-world applications in public spaces and personal safety devices. This work represents a significant advancement in accessible alert systems, extending situational awareness tools to the hearing-impaired and contributing to broader public safety.