Mind the data gap(s): Investigating power in speech and language datasets