Automatic Speech Recognition models require large amount of speech data for training, and the collection of such data often leads to privacy concerns. Federated learning has been widely used and is considered to be an effective decentralized tech- nique by collaboratively learning a shared prediction model while keeping the data local on different clients devices. How- ever, the limited computation and communication resources on clients devices present practical difficulties for large models. To overcome such challenges, we propose Federated Pruning to train a reduced model under the federated setting, while main- taining similar performance compared to the full model. More- over, the vast amount of clients data can also be leveraged to im- prove the pruning results compared to centralized training. We explore different pruning schemes and provide empirical evi- dence of the effectiveness of our methods.