Improving speaker turn embedding by crossmodal transfer learning from face embedding

Re-assign community