Learning electric guitar and toying with the idea of making a trainer game/app, primarily for practicing chords. Realtime chord detection is not too difficult for a single person to implement, but at the same time it is not a commodity technology either.
I'm using Rocksmith and Yousician (the free version), but I think there's room to improve over what both of them provide for practice / exercises. For learning and performing songs they're definitely the best due to the sheer amount of content they have.
I'm very early in my learning journey so I'm not concerned that developing a prototype version I could use for myself will take too long (it could be that I'm underestimating the difficulty of the problem). I was already able to slap together most of the "glue" code in Unity in a couple hours, and given my signal processing background I have solid ideas for how to approach the difficult algorithmic parts.
Since you have a signal processing background... what would be super cool is if there was a tool that could analyze all the chords in a song file (e.g., mp3) and magically create a Rocksmith/Guitar Hero style "racetrack" for practicing. DLC can get expensive quite fast, and songs you want to play may not always be available without modding/hacks/cracks, etc.
I'll keep this in mind! It is a tremendously more difficult problem though due to all the other instruments (including voice), distortion, etc. in real music tracks compared to a clean guitar signal. This is when you would have to break out the neural networks. Machine learning happens to be my day job so it's not impossible! But for now I'm keeping my target scope small :)
