In this paper, we propose a meta-universe-based emergency broadcasting system that integrates digital twin, multi-modal interaction and distributed rendering technologies to address the deficiencies of traditional systems in spatial perception and contextual interaction. The proposed system features a four-layer architecture integrating environment modeling, behavior perception, virtual-reality linkage, and user interaction. By employing digital twin, multi-modal interaction, and distributed rendering technologies, the system achieves sub-meter spatial localization, dynamic path planning, and real-time cognitive load optimization. Experimental results demonstrate that the accuracy of the system’s information reception is 42% higher than that of the traditional mode, and the emergency response time is shortened to 3.2 s; the multi-modal interaction (AR/VR + haptic feedback) reduces the orientation judgment error by 65.9% and improves the response accuracy of the hearing-impaired and visually-impaired groups by 2.17–2.71 times. The study provides a novel digital paradigm for public safety and advances the practical application of metaverse technology in disaster management.