This study presents a new model order reduction technique applied to room acoustic simulations using a high-order numerical scheme based on the spectral element method. The goal is to efficiently simulate iterative design processes in room acoustics, where the room acoustics with different boundary absorption properties are evaluated much quickly without solving the full wave equation. The wave-based methods are highly accurate but expensive when simulating high-frequencies and long simulation times in large and complex geometries, which are needed for industrial applications. Hence, it remains a key challenge to improve the computational performance of room acoustics simulations with different boundary material possibilities. With model order reduction, the boundary condition can be parametrized in the subsequent modelling. This allows to reduce the dimensionality of a wave-based simulator without compromising the transient wave propagation accuracy while enabling significant reductions of computational costs after initial training using proper orthogonal decomposition. We provide evidence that the combination of the high-order numerical scheme with model order reduction has significant potential value for building acoustics in terms of computational efficiency and accuracy.