ScreamSense: Enhancing Emergency Response with Deep Learning

With cities growing rapidly, there's an increasing need for smart surveillance systems that can help ensure public safety. Traditional CCTV setups depend heavily on human operators, which often causes delays in detecting emergencies. This paper presents a real-time scream detection system that uses machine learning and deep learning to recognize distress sounds in audio recordings. The system combines a Convolutional Neural Network (CNN) with a Long Short-Term Memory (LSTM) model, trained on a labeled dataset of human screams and background noises. It uses features like MFCCs and spectrograms to accurately detect screams, achieving around 92% classification accuracy. For real-time performance, the system integrates PyAudio for audio capture and a user-friendly interface built with Kivy/KivyMD, making it suitable for deployment in public areas, hospitals, and smart surveillance setups. The model performs well even in noisy environments, with a precision of 0.91 and a recall of 0.94. This work provides a scalable and responsive solution that balances safety with privacy concerns, aiming to improve emergency response systems.