The rapid progress in technological developments of small Unmanned Aircraft Systems (sUAS) or simply "drones" has produced a significant proliferation of this technology. From multinational businesses to drone enthusiasts, such a technology can offer a wide range of possibilities, i.e., commercial services, security, and environmental applications, while placing new demands in the already-congested civil airspace. Noise emission is a key factor that is being addressed with high-fidelity computational fluid dynamics (CFD) and aeroacoustics (CAA) techniques. However, due to uncertainties of flow conditions, wide ranges of propellers' speed variations, and different payload requirements, a complete numerical prediction varying such parameters is unfeasible. In this study, a machine learning-based approach is proposed in combination with high-fidelity CFD and CAA techniques to predict drone noise emission given a wide variation of payloads or propellers' speeds. The transient CFD computations are calculated using a time-marching LES simulation with a WALE sub-grid scale. In contrast, the acoustic propagation is predicted using a finite element method in the frequency domain. Finally, the machine learning strategy is presented in the context of fulfilling two goals: (i) real-time noise prediction of drone systems; and (ii) determination of propeller's rotation speeds leading to a noise prediction matching experimental data.