Accurate prediction of Energy Decay Curves (EDCs) plays a crucial role in analyzing the acoustical characteristics of the room and estimating the acoustic parameters. This study presents a deep learning-based framework for predicting EDCs from room geometry and material properties. Room configurations, including dimensions (length, width, height), and source-receiver positions, are generated using the Pyroomacoustics library to ensure physically-based acoustic simulations. Material absorption coefficients are obtained from measured data sets to enhance the realism of the input conditions. The corresponding EDCs and room-impulse responses (RIRs) are computed within the simulation environment, capturing detailed temporal decay behavior. The input features are normalized and structured for sequential learning, enabling the model to capture the temporal dynamics of the energy decay process.
Furthermore, key acoustic parameters, such as reverberation time (RT20), early decay time (EDT), clarity (C50, C80), and definition (D50, D80), are extracted from both predicted and simulated EDCs to validate model performance. The evaluation results demonstrate strong agreement between the predicted and simulated EDCs, which confirms the generalization of the model in diverse room scenarios. This framework offers an efficient, data-driven approach for predicting room acoustic behavior, with promising applications in architectural design, auralization, and real-time acoustic modeling.
Step by step instructions on How to use the Repository