I'm pretty sure the speech recognition for those transcripts happens in the cloud (necessitating that the raw audio be uploaded), unless the devices have gotten

I'm pretty sure the speech recognition for those transcripts happens in the cloud (necessitating that the raw audio be uploaded), unless the devices have gotten substantially better since I last looked.