Improving on-device speaker verification using federated learning with privacy