Automatic speaker recognition using low resources: Experiments in feature reduction and learning