Class token and knowledge distillation for multi-head self-attention speaker verification systems | Publicación