Of course, some special stuff like partial allocation and hence partial exhaustion of MT-32 is known to be emulated inaccurately. Notably, available partials are allocated in munt one-by-one (unless a pair is required for a ring modulating timbre). The control ROM code in contrast operates with partial pairs due to peculiarities of the LA32 implementation.

Also note, the partial allocation in the emulation engine is rather closer to the behaviour of new-gen units than old-gen ones. Because if this is about sound dropouts (or additional hardware noise of the old-gen units), it doesn't seem to be a noble goal to emulate accurately...

Still, it must be interesting to know the details, as it may very well be a bug in another area.

sergm wrote on 2023-04-06, 17:47:

Still, it must be interesting to know the details, as it may very well be a bug in another area.

As I said, I can provide the MIDI files (one initialization, one music), and a hardware recording if you want it. Should I attach the files here or would somewhere else like Github or Sourceforge be more convenient?

sergm wrote on 2023-04-06, 17:47:

Because if this is about sound dropouts (or additional hardware noise of the old-gen units), it doesn't seem to be a noble goal to emulate accurately...

I agree accurate emulation of the hardware noise is not worthwhile. But could it not be argued that correctly emulating sound dropouts might be worthwhile? This song *sounds different* on my 1.07 MT-32 than it does on MUNT (and presumably later hardware as well)

(I'm not suggesting or requesting that you go and implement this. I'm merely raising the argument for debate's sake)

Well, I have no specific requirement wrt. file sharing, so please feel free to do it as you see best 😀

Speaking of emulation accuracy, munt is not an emulator per se, so we do not expect it to approach 100% accuracy. There are many technical problems on the way to perfect reimplementation, but that is probably not needed anyway. What makes it even trickier to implement is the fact that the original hardware itself may sound differently from time to time. Still, when a reasonable effort improves the emulation accuracy, I don't mind emulating hardware noise too (in fact, that's what the analogue path emulation for). We have a number of options already for trading off accuracy vs. cleaner sound.

Here's the troublesome MIDI. This is a recording I made almost a decade ago. When I play this on my MT-32 *today*, I get a slightly different result than this, for some reason that eludes me, but it's still majorly different than MUNTs output.

OK it does sound like a difference in the partial allocation algorithms between the new-gen and old-gen units. Let's see what we can do without rewriting the whole thing 😀