A fundamental challenge in electron microscopy is bridging the gap between high-accuracy theoretical calculations and experimental observations. While density functional theory (DFT) provides precise atomistic calculations, it cannot handle the large-scale systems typically studied in experiments. Machine-learned potentials, trained on DFT data, offer a promising solution by delivering DFT-level accuracy while enabling calculations at significantly expanded spatiotemporal scales. In this work, we combine machine learning molecular dynamics (MLMD) with multislice simulations [1,2] to compute phonon spectra in complex materials at experimentally relevant scales. We then directly compare these results with monochromated electron energy-loss spectroscopy (EELS) measurements. To demonstrate the power of this approach, we investigate PbTiO3/SrTiO3 ferroelectric superlattices, where topological polarization vortices generate distinct vibrational signatures in EELS. Our MLMD simulations easily model large systems at scale, spanning 130 nm in length and containing 216,000 atoms, allowing us to not only reproduce the experimental EELS signatures but also investigate the underlying physical mechanisms. This work establishes a robust theoretical framework for advancing our understanding of vibrational EELS in complex crystalline systems, notably ferroelectric oxides and heterostructures where nanoscale structural distortions and topological phenomena play a critical role. By enabling accurate simulations of systems far larger and more disordered than those accessible to traditional DFT, our approach bridges the gap between theory and experiment, offering new insights into vibrational behavior at the atomic scale [3].