Improving Speaker Turn Embedding by Crossmodal Transfer Learning from Face Embedding | IEEE Conference Publication | IEEE Xplore