I think there is still alot of very interesting and uderexplored territory in the area of audio user interfaces, and I hope there is more interest in this now that the novelty of screens is finally wearing off.

For example there are spatial audio libraries that can make sounds apparently originate from any location for a user wearing headphones. Why not use that to create a spatial user interface?

Speech-based interfaces are of course very important, but i feel like there could be much faster ways of doing "window management" or diving through standard menus by using synthesized sounds for feedback. This would be attempting to go a step beyond just playing a sample or sound effect, and conveying more detailed information with different aspects of sound texture.

One of the biggest advantages of screens is that they also act like short-term memory. I don't need to remember where I am in a document because the scroll bar holds that information, for example. I feel like it should be possible to do an equivalent thing with audio-only without it becoming overwhelming. Our ears can take in about as much information as our eyes, and we can also "switch focus" and ignore background sounds when not immediately relevant. I don't know of any projects that seriously attempt to do this, unfortunately.