Federated Learning for Speech Recognition: Revisiting Current Trends Towards Large-Scale ASR