Indoor positioning is one of the most important tasks during disaster relief. As software technology evolves rapidly, various applications based on building information modeling and virtual reality have been utilized to simulate the three-dimensional scenes and sound effects of real-world buildings. For example, characters roaming in the virtual world can perceive sound absorption, scattering, transmission, and distance features. The purpose of this study is to construct the virtual replica of a building space, analyze the sound reception data of each designated point, and use the deep learning algorithm to identify the corresponding indoor position. In addition, although modern deep learning algorithms can produce satisfactory predictions, they may take longer time to reach convergence, which is not feasible during disaster relief. Thus, adjustment of algorithm parameters to balance the trade-off between model accuracy and training time is discussed, followed by model limitations and future directions.