A method of predictive simulation of flow-induced noise using computational fluid dynamics has been developed. The goal for the developed method was application in the vehicle development process, and the target of the research was therefore set as balancing the realization of a practical level of predictive accuracy and a practical computation time. In order to simulate flow-induced noise, it is necessary to compute detailed eddy flows and changes in the density of the air. In the research discussed in this paper, the occurrence or non-occurrence of flow-induced noise was predicted by conducting unsteady compressible flow calculation using large eddy simulation, a type of turbulence model. The target flow-induced noise for prediction was narrow-band noise, a type of noise in which sound increases in specific frequency ranges. Assuming the area of generation of flow-induced noise to be the exhaust pipe, including the complex shape of the muffler, predictive accuracy was verified under conditions that modeled measurements in steady-state flow test equipment. In order to reduce computation time while maintaining predictive accuracy, calculation methods were combined and computation times compared. This made it possible to reduce computation time by 61% against that in the initial stage of development of the method. In addition, a study utilizing multiple exhaust pipe shapes indicated that a correlation existed between the number of computation cells and computation time. The developed method has made it possible to predict the occurrence of narrow-band noise within a practical computation time.