In defence of metric learning for speaker recognition - View it on GitHub