I haven't gotten a chance to look at it too much yet, but what is special about the AZ music engine?
Is it just MML? Or is there some extra secret parts added in??
There's actually nothing special about the Air Zonk engine. It's kinda simple in design. The real engine is actually pretty
inefficient. It runs the whole thing off the TIMER instead of VBLANK. The engine as a whole eats up a whopping 20% cpu resource from what I clocked it at. When it should only be around 7-8%. It has a compressed samples (that have been preprocessed to make them more compression friendly), but the compression doesn't save much space and adds most of the extra overhead.
Anyway, the engine is a command string system or MML as it's commonly referred to. It's based/similar on mml structure, but it is its own design. The command string structure (mml) is pretty flexible in execution - more so than a typical tracker format. A single entry can contain up to 256 bytes of commands per channel, but I doubt you'll ever need that many. For example, you can reload envelope registers on every note if you wanted. I was documenting and decoding the Air Zonk engine for myself and someone who was interested in doing remixed versions. I had pretty much the whole thing figured out, so I decided to write a player for the song data as an exercise.
HuC has the same MML player that Mkit has, it just happened to get disabled in the latest release. If you look through startup.asm, you can see where it's accidentally commented out. Dave Shadoff wrote the player based on the Bios player, but he stated that it was untested. You're probably better off writing your own anyway. It's fairly easy to write such an engine. Writing the converter ascii format to bin file is another story.
http://pcedev.net/docs/Air_Zonk/example.txt <- an example of the syntax(ascii version) I'm working on for a converter.