Just reading a Wired News article on the need for assistive technologies. Here's a thought:

Define a protocol using bluetooth, or some other proximity-based system, such that a "personal client" device can signal to nearby machines that it requires textual interaction (thinking here primarily of displays for the blind). The machines have a very cheap transmitter with the ability to recognise the personal client's signal and return the text from the screen (as text).

The personal client, then, is a far more expensive device, with the ability to do text-to-speech processing. The text transmitted from the machine could be far more detailed and vision-impaired-friendly than the GUI interface (click the third button down on the side of the screen to accept the transaction, click the fourth button down on the side of the screen to correct the transaction... versus "Correction ->").

Basic principle is that you associate the expensive part of the technology with the person who actually needs it, then use the least expensive mechanism possible to provide information to that device.

The idea also doesn't require that every device be made aware of it. What it does is it makes it a minimal outlay to accomodate a vision-impaired citizen, resident, employee or customer. You'd imagine starting with major public devices (e.g. bank teller machines). When the cost comes down sufficiently, it then becomes feasible to build it into the cost of major appliances (washers, dryers).

You want to keep the protocol as simple as possible, and maybe it would be better to provide contact-based communications (rather than proximity-based) mechanism to reduce the risk of eavesdropping (particularly wrt bank transactions))... but you'd want the wireless options for more mundane "what's going on" type communications.