Google Voice does VM-to-text synthesis, but I don't think it's behavior is along the exact lines of what you described.