To clarify, the exhaustion doctrine allows the owner of the licensed work (ebook, etc) to access/display/use the work except as disallowed by the license.

This includes viewing it on a computer monitor. It's irrelevant that the computer monitor maker does not have a license to the work, since the viewer does.

In the case of the audiobook speech-to-text argument: the licensee has a license to the audiobook but not the prose version of the work, which is a legally distinct creation under many decades of international and domestic IP law.

So Amazon or the customer needs to have the license to the prose work for speech-to-text to be defensible, but neither do.

I am not convinced by this. Amazon does not at any point convey a copy of the written work to the customer. At no point does such a copy exist, nor can it conceivably be constructed from the technology Amazon are supplying to their customers.

This is not (if I understand correctly) a PDF document they can download and read, it's a technology that allows the words that are currently being spoken to be seen in written text.

At no time does a legally significant quantity of the written work get conveyed to the customer. Amazon will just ask the publishers to show where this alleged copy of the textual work resides in their system. They won't be able to do so, as it doesn't exist there.

This does seem to be what they are in fact arguing. And I think they will prevail. I would not expect them to prevail if they offered a pdf of the book or if their software actually contained a copy of the original text that was being conveyed to the customer.