Improving Fairness in Speaker Recognition