There's just one problem: privacy. I don't have that when using voice input. Writing won't decline in favor of voice input until we find a way to make it private. Tangential to this, smart glasses won't replace smartphones either for the same reason.

Viable on-device speech recognition is becoming reality. Apple is apparently processing some speech recognition on-device now: https://www.engadget.com/ios-15-siri-on-device-app-privacy-1...

So far there aren't any particularly good open source voice recognition models though, in large part due to a lack of training data. You can (and should!) contribute to Common Voice to help change that: https://commonvoice.mozilla.org/en

No matter what software you use, everyone standing next to you can hear the entire message you're composing.

Make that privacy and accents. There are too many regions where choice recognition just won't work without essentially building new voice models. And that's before we get to people using their second/third/... language.

I think in the later books in the Enders Game series (Speaker for the Dead, in particular), Ender speaks to his AI companion via subvocalization.

So perhaps a little sensor on or in our neck will allow us to speak to computers silently.

may not work so well if you want to write something in an area where other people are talking or with a TV/podcast in the background.