Real Time Spatial Sound Scene Analysis-AlertNet

This project introduces a groundbreaking approach to enhance the accuracy of acoustic scene classification and ensures user safety within music player applications. It provides a combined model of deep learning by merging Convolutional Neural Networks (CNN) with Bidirectional Long Short-Term Memory (Bi-LSTM).The model uses Mel-Frequency Cepstral Coefficients (MFCC) and Gammatone Frequency Cepstral Coefficients (GFCC) features as feature extractors, and processes these features to predict scenes.Furthermore, this paper integrates the advanced scene classification model into music player applications. Users can enjoy their favourite music while the system continuously analyses the audio environment. When a potentially dangerous situation is detected, such as sirens, alarms, or other alerting sounds, the music playback is momentarily attenuated or paused, and users are promptly alerted to pay attention to their surroundings. Performance evaluation demonstrates $97 \%$ accuracy in a noisy environment and $98 \%$ accuracy in clean environment.The CNN-Bi-LSTM hybrid model, named AlertNet, excels in capturing both fine-grained and contextual audio information. Its seamless integration into music players enhances safety and user experience. Finally, it exemplifies the fusion of cutting-edge deep learning techniques with everyday technology to ensure safety and situational awareness for users, especially in scenarios where distractions are common.