There are two areas which I think are actively being researched but are interesting and could probably be improved by a masters level project. One is the idea

There are two areas which I think are actively being researched but are interesting and could probably be improved by a masters level project. One is the idea of separating multiple audio tracks. For example, if you have a MIDI file of two instruments, exported to WAV, it should be possible to train an AI to produce two WAV files that sound exactly like the individual instruments being played on their own. (Last time I heard a demo of this, it seemed like the AI hadn't built a good model for what the individual instruments should sound like, even though it should be possible to infinitely generate the training data). The other is an AI for turning hand-written mathematical notation into a proper digital form (like LaTeX). Ideally you'd be able to give the AI a feed of a lecture, where notation was written on a chalkboard, and the output of the AI should be valid mathematical expressions. Bonus points if you can validate that the mathematical steps are correct (possibly by finding them in an existing corpus of mathematical steps).