It might be practically necessary, though, if they need to have a huge corpus of data in memory in order to perform their voice recognition.