As you might have guessed, many audio-only UIs have been developed for blind people over the decades. One example of an audio-only PDA (from the time when the term "PDA" was still widely used) was the LevelStar Icon. (Disclosure: I did some contract work for LevelStar in 2005-2006 and was friends with the lead developer.) After some digging I found an old audio presentation about the LevelStar Icon here:

http://www.vibug.org/audio/vbg0407.mp3

I started developing a blind computing system in the late 80s based on a BBC Micro and a very early speech synthesizer.

It never went anywhere, and I'm no longer in contact with the blind friend who was helping me, but I often wonder if today's solutions such as Apple's Voiceover aren't the ideal solution.

They seem to be trying to adapt the GUI for the blind, rather than trying to build a text/voice based system from scratch, like I was doing.

Siri / HomePods and similar appliances are getting there, I suppose, but they still feel like computing accessories, rather than fully fledged computers themselves.

As a legally blind developer who has worked on accessibility for a while (see my HN profile for more), I'm inclined to agree with you. We use screen readers like VoiceOver because, practically, we often need to access mainstream applications. But I think we tend to get too dogmatic about going mainstream, to the point that some of us are disdainful toward systems designed specifically for blind people, particularly if they involve custom hardware. I've thought about developing a Linux desktop environment with a shell that's designed specifically for TTS, but that also includes a more conventional screen reader for running GUI applications.

Do you remember which speech synthesizer you were working with in the 80s?

Yes it was a software synthesizer called "Speech!" by Superior Software, blown into a ROM.

Ah, must have been a BBC Micro-only thing, because I never heard about it.

It certainly was, yes. The BBC Micro had an optional hardware speech synth you could add, but it was very inflexible, limited to a fairly small set of pre-defined words. "Speech!" could render arbitrary phonemes which make it a lot more flexible.

Flexible enough to say the word "Citadel!" for example. :)

</nostalgia>

You can try it in your browser; Go to https://bbc.godbolt.org/?&disc1=sth%3ASuperior%2FSpeech.zip, then copy and paste this to the paste thing at the top right:

SPEECH SAY Hello *SAY How are you?

Thanks for the link. :)

My comment was intended as an incredibly subtle reference to the trivia that the BBC Micro game "Citadel" 'famously' had speech generated by "Speech!" in its intro screen (but appreciate you taking the time to share the link :) ) as shown here : https://youtu.be/Hu5vu9SgGZI?t=44

For anyone else wanting to try out the emulator link in the comment above, HN messed up the formatting, so the asterisk prefixes were missed on the first two commands:

    *SPEECH

    *SAY Hello

    *SAY How are you?

I'm very curious about your thoughts. How would you make a shell designed for TTS?

There's emacspeak.

I remember ADRIANE being included in the Knoppix ISOs back in the day. You might find it interesting: http://www.knopper.net/knoppix-adriane/index-en.html

It's less comprehensive but depending on your definition devices like Amazons Echo or Googles Home devices could count as audio only OSes