Amazon is not fundamentally offering a textual copy of the work. If they were doing so they would be violating the license. Instead they are offering a technology to transform the (possibly flawed performance of) the work using a tool (that is itself possibly flawed). That is not the same thing as distributing a textual copy of the original work.
A computer is a device on which the work is licensed to be distributed, and it happens to have the capability to transform the work, just as a volume knob or a graphic equalizer would.
If Amazon can't offer this technology because they are licensed to distribute the work, what is to stop someone selling such technology who is not licensed to distribute the work? The effect is the same, precisely because it is a general purpose computing device with sophisticated functionality. And yet I don't see publishers getting far with a claim that all manufacturers of speech-to-text devices (think of deaf people) should pay them because the device could be used to transform their works.
Of course I know this is not how any publisher sees their work. E.g. you need a different license to perform a work than you need to modify or make a video production of a work (mechanical vs sync license, etc.)