Phonopy implemented this relatively recently. It would be nice to be able to remove some internal code and depend on phonopy for this as well.