Oh right, those worked by calling up someone who also had a teletyper who would then handle the speech part of whatever call they were making. I forgot it was a human doing the “translate from typed to spoken and back” and assumed there was a text-to-speech and voice-to-text in the loop.