I like this proposal and would modify it with a very robust interface I saw at a music exhibit at a children's museum (it might have been Exploris in Raleigh, NC), but I'm not sure). Instead of buttons or any moving part that kids are almost certain to break quickly, use beanbags with RFID chips. Different colored beanbags might represent different sounds, and you could distribute them along a long rubberized pad (representing a time axis). Little hands have no difficulty picking up and dropping soft beanbags. If the long pad matched a long display, then the placement of a beanbag would correspond to the sound and light emitted from the display at that point in time. Lots of kids could place lots of colored beanbags along the pad and interact with the resulting sound loops very easily, and at very low risk of breaking the UI.

I was struggling to remember my login so that I could post the same info from Cool Tools - I'm glad you did! I highly trust the recommendations that are made on that site.