As a random observer I always thought that a linear text based OS would be good for blind people, basically what the CLI is but with more metadata, a tree structure so things not of interest could be skipped, eg a long list of filenames. A CLI with folding.
Actually GUIs or websites, if done right, can be way better than anything that is text based. They include much more semantic information that a screen reader can use. For example, on websites, we have special shortcut keys to i.e. jump to next heading, table, landmark etc. We can even navigate within tables. This makes using most web/electron/chromium apps pretty easy. Most blind people prefer traditional win32 guis, but I actually don't. Problems arise when semantic information isn't provided, as in a web developer making the text whatever color is fashionable now and adding an onClick instead of using <button>. I don't see how a text-based OS would be different from that.
Web UIs can be used with my screen reader's search functionality and quick navigation keys. For example, in Spotify, which uses a web UI, I can do ctrl+ins+f, type rock, press enter and I'm focused on the Rock playlist. Similarly, in Skype, I can search for "audio call" etc. Some elements can also be reached pretty quickly with navigation keys. I think it's much easier to screw up a win32 gui than a web app. The non-web paets of Itunes are a good example. They're accessible, but all objects need to be reached by pressing tab, and there's a lot of objects. There's no semantic structure and no way to provide one. A website-like document would work better in this case.
Many years ago, I was looking into emacspeak [1], which seems like it could work for this. (I thought it would be a cool way to interact with a wearable computer; I didn't get much farther than those thoughts)
Emacspeak is something I always wanted to try, but keep putting it off for later. Emacs is a powerful tool in itself, and the way Emacspeak integrates with it is unprecedented. It's not just a screen reader getting stuff from OS APIs, it knows a lot about internals of most Emacs features so it can make the speech output as efficient as possible. It also has some cool features like sounds in stereo/3d, changing voice parameters for different text styles etc.
