You don't. The descriptors are used by the visually impaired. The goal is to keep image/video only.