I Used the Web for a Day Using a Screen Reader
smashingmagazine.com
smashingmagazine.com
That being said, this was an interesting read.
It's a losing battle trying to get every technology and web service to put in all the right aria tags - imagine imgur for example - there is no way you could make those all those animated gifs and mspaint memes properly accessible - instead, a neural network should look at the screen like a human does, and describe it.
The screen reader should say "There is a gif of a squirrel falling off a shelf, and it has a share and an upload button"
But, as you said, OCR isn't perfect. I'm not surprised that it still has inaccuracies when recognizing code in particular. Also, computer vision takes a lot of computing power. I'm going to be sad if someday it ends up taking a powerful GPU, possibly running on a remote server, to enable a screen reader to do the things that a 90s computer could do without breaking a sweat via programmatic accessibility APIs.
Not true for Windows screen readers and VoiceOver for iOS. There's a widely implemented feature called audio ducking, where the OS turns down the audio of other applications while the screen reader is speaking. So the music may be quietly in the background, but it's still there. Does Mac not have this feature?
I think that accessibility can improve considerably with HTML5 and CSS grid. There is no need for pages to be a sea of divs, the document structure can be dome correctly so the page looks good with no styling, then the CSS grid layout engine can be used to make it look pretty for the sighted folk.
I think that document outline is really important, and to ask yourself the question of how you would present something if it was typed up, Word document style or like a formal letter.
There is no excuse for developers to not be using the correct tags, so that means header, main and footer, with articles and asides. The nav tag can be used for anything that is navigation and structured in a list. Pictures with figures and figure captions can be used too. All these elements can be targeted with CSS to be styled, there is no need to have documents have 17 nested divs each of them with many meaningless class tags when there are perfectly good elements to style. You can take a document and get the structure right and then use the elements as a skeleton in SCSS to style it. If structured properly these need not go '17 deep'. I think that if people had div-less and class-less markup and were used to styling it that way with SCSS generating the CSS then it would seem absurd to go to the unmaintainable sea of divs and class tags that passes for normal today.
There does need to be more work done with HTML5 to get the perfect outline, I do use outline tools already but that is a fine art.
Back to the article, the BBC do actually have people that use screen readers, I met one recently and how he used his iphone was a sight to behold, just playing with it more like how Jimi Hendrix played the guitar.
Since the author works for the BBC and with the BBC being so vast he might not know the screen reader community that work there.
I have suggested to people I have worked with that we get an actual partially sighted person to test web pages for accessibility. I got a 'why would you want to do that' attitude. However, I think that this is a service that should be offered to clients.
To be clear, though, there's another abuse of divs that is a problem: using a div with an onclick event handler and no ARIA role where a simple button would do. Don't do that. I'd suggest using an actual button tag if possible, so you get both keyboard and screen reader accessibility for free. But if that doesn't work with your CSS for whatever reason, at least put role="button" on that div.
Just today I tried using a screen reader to surf the web, for the first time ever. It's a completely different experience, requiring a lot of patience. Many websites were too difficult and frustrating to navigate. And I've been working in web development for years.. It's a shame that I hadn't deeply considered accessibility, and I've vowed to become more familiar with it, to incorporate and apply it at work.
(and the OCR part could probably easily be matched with the source code to get perfect error correction)