We developed a method to filter phonons from molecular dynamics (MD) simulations based on a density functional tight binding approach. We are able to extract phonon frequencies, polarizations, and q-vectors. Our results for silicon bulk and silicon nanowires show good agreement with experimental data and exceed the accuracy and applicability of the Hesse matrix approach. In particular we are able to determine phonon occupations and temperature effects. Furthermore we filtered Raman, Infrared, and silent modes from MD trajectories via symmetry analysis. This enables us to predict Raman and Infrared spectra of silicon nanowires and explain their origin in terms of phonon polarizations.