They still needed a human to process the files, and another human to provide the voice model.

I mean, the voice files are specifically for a robot, they could have not had a human voice involved at all and eliminated one human from the job pipeline regardless by using FM synthesis or something, which has been around since before the 1990s.