The more scalable way would be to have reconfigurable wake word hardware, then push target words, then send back analytics instead of audio.

Especially since Amazon already has advertising profiles on users, so could generate monetizable target word lists per-user.

> So record everything and upload it along with the legit uploading. Just hide it in there. Seems trivial.

You really think this would go unnoticed?

> You could even convert to text first to make the packet super tiny.

This entire debate came up in the context of them having to upload audio clips for server side processing.

So which is it? They’re uploading Audi clips? Or they’re converting everything to text?