Abstract Conventional rock physics experiments and well logging analyses face challenges in deep carbonate formation containing meter-scale cavity-fracture system, since it is unfeasible to extract representative samples from targeted formation. In this study, differing from the conventional digital rock models at the core scale, we construct a digital rock model from realistic outcrop images with meter-scale heterogeneity of cavities and fractures. Using a dynamic stress–strain simulation, we investigate how reservoir type, fracture density and spacing, cavity porosity and size, fluid saturation, and connectivity influence wave dispersion and attenuation. By selecting the simulation frequency of 0.3 MHz approximately corresponding to the field seismic measuring frequency of 30 Hz, the simulation results can reflect the realistic seismic velocity and attenuation characteristics of deep carbonate reservoir rocks. We find that the P-wave velocity decreases linearly with increasing fracture and cavity porosity. The P-wave attenuation is sensitive to variations in fracture density but shows limited sensitivity to changes in cavity porosity and size. Fractures with narrower spacing result in higher attenuation when fracture density remains constant. Deep carbonate rocks with cavity-fracture systems saturated with two immiscible fluids display enhanced attenuation compared to fully saturated conditions. The velocity demonstrates an almost linear increase with water saturation, while attenuation initially increases before decreasing at a peak water saturation of 70%–80%. The connectivity between fractures and cavities significantly influences the attenuation behavior. Our simulation results provide valuable guidance for seismic data processing and interpretation of deep carbonate reservoir rocks with cavity-fracture systems.