yeah. with FL it should be possible to make sense out of all data that is distributed across devices without ever having to move it to a central location (i.e. collect it). In the case of speech data, users participating in a federated setting would likely come from different backgrounds, which could be reflected in their accent or use of language.