Some screen readers can do OCR on a specific part of the screen (e.g. an unlabeled image) on request. But while OCR is useful for getting text out of an image, most implementations can't discern the structure of a UI, e.g. which part is a button, which part is an edit box, etc. Also, OCR is typically done just once on-demand, not continuously as the screen changes.
However, VoiceOver for iOS has a new feature called screen recognition, which is exciting because it overcomes these limitations and provides some level of access to applications that are otherwise inaccessible. Hopefully other platforms will catch up.
Even then, true screen reading will be much more CPU-intensive than what screen readers currently do. And anyway, it's not here yet, except on iOS. So I will continue to warn developers away from toolkits that are inaccessible, in hopes that some blind person somewhere will be spared the pain of being blocked from doing a task because of an inaccessible application.
However, VoiceOver for iOS has a new feature called screen recognition, which is exciting because it overcomes these limitations and provides some level of access to applications that are otherwise inaccessible. Hopefully other platforms will catch up.
Even then, true screen reading will be much more CPU-intensive than what screen readers currently do. And anyway, it's not here yet, except on iOS. So I will continue to warn developers away from toolkits that are inaccessible, in hopes that some blind person somewhere will be spared the pain of being blocked from doing a task because of an inaccessible application.