The exact mechanism of endothermic singlet fission in crystalline polyacene remains to be clarified. It has been elusive whether the excess energy of vibrational hot states and the upper branch of Davydov splitting is important for the energy compensation. Here, we probe the excited-state specified singlet fission dynamics in tetracene single crystals by polarization-dependent two-dimensional electronic spectroscopy (2DES). While a major spectral transfer with a characteristic lifetime of 86 ps is observed to be largely independent of the excitation energy due to formation of the spatially separated triplet pairs ((1)(T center dot center dot center dot T)), the excitation-energy dependent subpicosecond dynamics show marked differences for different states probed, implying the possible involvement of a coherently formed triplet pair state ((1)(TT)). Analysis of coherent vibrational modes suggests the coupling to high energy modes may offset the energy difference between singlet and triplet pair states. Moreover, the beating map of the low frequency mode indicates a vibrational hot state violating the aggregation behavior of Davydov exciton, which can be explained as a resonance of the (1)(TT) state. These results suggest that the coherent vibronic mixing between local excitation and triplet pair states is essential for the singlet fission dynamics in molecule aggregates.