Abstract: Iterative self-training, or iterative pseudo-labeling (IPL)—using an improved model from the current iteration to provide pseudo-labels for the next iteration ... and established i-vector ...