Text-independent speaker recognition based on adaptive course learning loss and deep residual network.