Advanced Deep Learning Framework for Decoding Emotional States Through Temporal and Spatial Analysis of Ambient Soundscapes

The paper focuses on creating an efficient system that can predict the subject's emotional state through analyzing sound environment. Thus, we use open-source audio datasets of various environments including cafes, parks, and libraries that are labelled as calm, anxious, and joyful. Audio traits are described by using Feature extraction techniques in librosa Python packages inclusive of Mel-frequency cepstral coefficients (MFCCs) as well as the spectral centroid. The characteristics obtained from the soundscape are then used to develop CNNs and RNNs for emotions recognition from soundscape characteristics. The proposed models are evaluated on the hold-out data to ascertain the reliability, and the accuracy of the emotions predicted. The findings provide information about acoustic features associated with certain emotions. It has the possibility of monitoring and identifying subjective state 24/7 usage in smart homes, health care settings, and city planning.