Dynamical heterogeneities--strong fluctuations near the glass transition--are believed to be crucial to explain much of the glass transition phenomenology. One hypothesis for their origin is that they emerge from soft (Goldstone) modes associated with a broken continuous symmetry under time reparametrizations. To test this hypothesis, we use numerical simulation data to construct coarse grained observables and decompose their fluctuations into two transverse components associated with the postulated soft modes and a longitudinal component unrelated to them. We find that as temperature is lowered and time scales are increased, the time reparametrization fluctuations become increasingly dominant, and that their correlation volumes grow together with those of the dynamical heterogeneities, while the correlation volumes for longitudinal fluctuations remain small.