Accurate prediction and optimization of the acoustic behavior of composite materials require modelling complex multiscale poro-elastic interactions, resonance phenomena, and anisotropic microstructures. Conventional experimental characterization and numerical simulations, such as finite element analysis, are computationally expensive and time-intensive, limiting their effectiveness for rapid acoustic material design. Artificial intelligence (AI) has recently emerged as a promising surrogate modelling approach capable of accelerating acoustic prediction and optimization. However, existing review studies on AI applications in acoustic materials largely provided descriptive overviews of algorithms without systematic quantitative benchmarking or integration of acoustic physics, making objective comparison of model performance and generalizability across composite systems difficult. This review addressed this gap by presenting a physics-integrated and quantitatively benchmarked framework for evaluating artificial intelligence models used in acoustic composite design. Machine learning techniques, including artificial neural networks, deep neural networks, convolutional neural networks, Gaussian process regression, ensemble learning approaches, physics-informed neural networks, and hybrid finite element machine learning architectures, are systematically analyzed using standardized evaluation metrics such as coefficient of determination (R²), root-mean-square error (RMSE), broadband sound absorption deviation, resonance prediction accuracy, and computational efficiency. In addition, generative modelling and optimization strategies, including variational autoencoders, generative adversarial networks, genetic algorithms, particle swarm optimization, Bayesian optimization, and NSGA-II, are examined for their effectiveness in acoustic performance. The review further identified key methodological challenges, including dataset scarcity, model interpretability, uncertainty quantification, and cross-material generalization. By integrating acoustic physics with machine learning evaluation, this study provided a reproducible benchmarking framework and practical guidelines for the development and optimization of next generation data-driven acoustic composite materials.
