For large eddy simulation, it is critical to choose the suitable turbulent inlet boundary condition as it significantly affects the calculated flow field. In this paper, the effect of different inlet boundary conditions, including random method (RAND), Lund method, and divergence-free synthetic eddies method (DFSEM), on the flow in a channel with a hump are investigated through large-eddy simulation. The simulation results are further compared with experimental data. It has been found that turbulence is nearly fully developed in the case based on the Lund method, not fully developed in the case based on DFSEM, and not developed in the case based on the RAND method. In the flow region before the hump, mean velocity profiles in the case applying the Lund method gradually fit the law of the wall as the main flow moves towards the hump, but the simulation results based on the RAND and DFSEM methods cannot fit the wall function. In the flow region after the hump, cases applying Lund and DFSEM methods could relative precisely predict the size of turbulent bubble and turbulent statistics profiles. Meanwhile, the case based on the RAND method cannot capture the positions of flow separation and re-attachment point and overestimates the turbulent bubble size. From this research, it could be found that different turbulent inflow generation methods have a manifested impact on the flow separation and re-attachment after the hump. If the coherent turbulence is maintained in the approach flow, even though turbulent intensity is not large enough, the simulation can still predict the flow separation and turbulent bubble size relative precisely.