[fsmv]

It's possible to do some voice transcription today but nobody has the budget or number of chips needed or bandwidth or edge device battery power to ingest billions of 24/7 audio streams and transcribe them in real time for ad serving.

On top of that they'd have to make it a conspiracy and make sure no engineers talk about it here on HN.

We can do those voice assistants only because everyone isn't using them simultaneously for 24/7 transcription.