Real-time room acoustic modeling is required in many situations, especially for interactive applications (augmented reality, virtual reality, video games) where plausible reverberation contributes to the sense of immersion. Since real-time audio rendering with interactive parameters can be computationally demanding, one of the ongoing concerns in the field is the design of more computationally efficient room acoustic models. On top of being all-around efficient, it is desirable for models to be “scalable”, meaning that one may choose to gain efficiency at the price of more approximate acoustic modeling, or vice-versa. Most existing models have limited scalability, with accurate models being computationally expensive, and real-time-capable models presenting approximations. For example, wave-based models offer higher physical accuracy than any other model class, but their computational complexity makes them currently unviable for real-time, broad-band, interactive applications. Geometrical acoustics, albeit more approximative, can be employed much more efficiently; however, they generally require the use of convolution, a process that can itself be computationally intensive. The need for convolution can be eliminated through the use of delay line networks, a class of room acoustic models, thus greatly reducing computational costs. The drawback of such delay networks is that they usually do not model any physical aspect of acoustics explicitly, aiming instead to replicate certain perceptual attributes of reverberation. The objective of this thesis is to design efficient methods for real-time room acoustic modeling, making use of delay network structures while also implicitly performing physical modeling of the desired acoustic space. A secondary objective of the thesis is to facilitate the designed methods’ efficiency/accuracy scaling, as defined above, such that a single model can meet the requirements of a wide variety of applications. These objectives are tackled through the investigation of existing delay network models with physically meaningful parameters (scattering delay networks, acoustic rendering networks). Said existing models make use of notions from geometrical acoustics, in particular the image source method and acoustic radiance transfer, to strengthen the connection with physical parameters. These delay networks are studied and extended, producing new models with improved scalability as well as enhanced ability to model arbitrary acoustic environments. The investigation culminates with the proposal of a novel delay network model based on the modal analysis of acoustic radiance transfer. The proposed model achieves a decrease in computational complexity of at least an order of magnitude with respect to ray-tracing, and offers considerable scalability with respect to both early and late reverberation modeling.