the mu100 instruments do improve a number of ff7 XG midi files, actually.

Oddly enough, it seems that game was kept in mind with the quality upgrades. Still more FIghting's distorted guitars are much improved, for example.

When they did the XG MIDI for FF7, they did it using the MU hardware and not the brand-new software, so it should always sound better with XG hardware from that time period. Once this gets a little more baked and fleshed out, it would be great to see YT vid recordings of the FF7 XG MIDIs using the better samples that have been dropped in and then, move on to some of the massive pile of XG MIDIs that were made over the years, using the more advanced hardware options.

test them out in s-mu2k.

Try mu-native first, then try mu-basic. you might be surprised. mu-basic map s for midis that soudn wrong withotu it. always has been, always will be. if it sounds great without it, you don't need it. 😀

here's still more fighting.

sw1000xg (it's a mu100 on a card) https://www.youtube.com/watch?v=OGLOqyya6Q4
real mu-2000ex https://www.youtube.com/watch?v=n8MGLBfrRAs

still, nothing beats the original hammond b3 slowdist organ sample from the PS1 original, though.

full soundtrack on sw1000xg

https://www.youtube.com/watch?v=71TxIvlenZ8
https://www.youtube.com/watch?v=zmLEPc1EE3U
https://www.youtube.com/watch?v=Y8hkwHXu2kk
https://www.youtube.com/watch?v=COB4n-R2nN8

see for yourself. 😀 this is what you will hear on s-mu2000xg

In case anyone hasn't been aware of it:

The S-YXG50 panel looks like this now and the engine supports polyphony 256. Just sayin'. And MU50-1000 can be loaded with it. Some better, some worse, but they do work.

Oh yeah, and there's a patcher which just needs the ROMs and syxg50.dll and you get ready-to-use VST2 dlls as a result. Single files, ROMs embedded, fully portable. It won't get any easier than that. 😀

Trelokk wrote on 2026-10-15, 20:00:

In case anyone hasn't been aware of it:

The S-YXG50 panel looks like this now and the engine supports polyphony 256. Just sayin'. And MU50-1000 can be loaded with it. Some better, some worse, but they do work.

Oh yeah, and there's a patcher which just needs the ROMs and syxg50.dll and you get ready-to-use VST2 dlls as a result. Single files, ROMs embedded, fully portable. It won't get any easier than that. 😀

I tried it yesterday, with all MU variants.
It is very user-friendly indeed. The Windows patcher requires Windows 10, Windows 7 won't do, But there is also a Linux one. Anyways these are pyinstaller executables. the core is python (v3.12+).
In my brief testing I noticed MU50 and 80 being closest sounding to my expectations, of certain midi tracks. But the others are not far off, when configured as MU basic.

gmaverick2k wrote on 2026-10-08, 15:26:

Got s-mu2000 working using clear instructions from a website

https://moondoldo.com/wordpress/?p=14483

So now that these converter projects are around anyone up for making a .TBL to .sf2 or .dls converter? For use with something like Spessasynth?

I've worked on a python script that converts the 1x and 2x element MU2000 ROM voices to Bitwig Sampler presets. These are very similar to .sf2 underneath. They sound like recognisable XG voices, but they're obviously quite raw sounding. I got as far as mapping loop and tail, AEG, FEG and ome PEG. There's quite a few challenges though because several aspects of XG/AVM2 parameters don't really have .sf2 equivalents so even though the sample and envelopes are identical, some of the character is not shaped properly. I did think it might be better to do what i did with MUXG2K and bake some of the AVM2 goodness into the samples where possible so it would make the encoding to .sf2 a little more accurate.

Wasn't the real secret sauce of XG, the DSP-ish stuff (processing commands and such) that was part of the XG MIDI standard and baked into the XG MIDI data?

For the time, what was unique was the PCM was source material for a complete subtractive synthesis engine that comes afterwards. Yamaha's digital filter at the time was also what allowed it to sound analogue-like filtering. This is why merely extracting the ROM waves is only a tiny part of AWM2 and why it's so difficult to replicate in software too.

unfortunately sf2 without extensions is incapable of handling gs or xg adequately.

.sf2 provides for 2 lfos per part. that much works. ut it also only provides 2 envelope genreators, oen of which can only be applied to volume.

Roland GS synths use three envelope generators. one for volume (TVA) one for filter (TVF) and one for pitch. XG gangs the volume and filter envelopes together. Neither can be put into sf2, because all envelope generators have independent defaults per patch.

AWM2 predates xg, and was used by the TG300 before it was used for XG.

That's not technically correct zaphod77 because AWM2 (hence XG/MU) voices do have independent per-element envelopes: iFEG (filter) and iPEG (pitch) which is the same count as GS, so there isn't a shared vol+filter EG as such. AEG is shared across all elements of a voice though, there isn't an iAEG to AWM2.
What you're probably meaning is real MU hardware / MAME / S-MU2000 keeps the iFEG and iPEG hidden and non-editable and running internally as kind of a baked-in-the-ROM and never exposes them at sample playback on the XG MIDI surface to the user. And it's the same for PEG. There's the user-editable global PEG (works like AEG across all element), but each element making that voice has it's own internal iPEG too.

This was a major reason for not going the MAME/SWP30 emulator route with MUXG2K and writing my own AWM2 engine in C++ native (took me 2 years developing it) was because I wanted user-editable control over more things within AWM2 than Yamaha exposed.

On the XG voices that contain them, you'll see that MUXG2K exposes them as editable envelope controls to the user as iPEG and iFEG on each element.

The reason for exposing them and making them editable in Realtime or saved to presets is because ultimately, MUXG2K will be made available to support MPE and MIDI 2.0 and these parameters then become part of the 5D control too. E.g you might have Slide controlling FEG cut-off, but Pressure iFEG 1 resonance and iFEG2 cut-off for example.

Exposed internal Yamaha PEG

Exposed internal Yamaha FEG (Element 1)

Exposed internal Yamaha FEG (Element 2)

Here's how much of XG / AWM2 actually maps to SoundFont 2

1XG / AWM2                   SF2 2.x equivalent        Fit
2--------------------------  ------------------------  -------
3Multiple elements / voice   Instrument zones          Direct
4Sample + loop               sample + sampleModes      Direct
5Root key / coarse+fine      overridingRootKey + tune  Direct
6Key-follow (pitch track)    scaleTuning               Partial
7Pitch EG                    modEnv -> pitch           Partial
8SFX scripted pitch (siren)  (none)                    None
9Vibrato (pitch LFO)         vibLFO -> pitch           Direct
10Amplitude EG                volEnv (DAHDSR)            Direct
11Low-pass cutoff / resonance initialFilterFc / Q       Partial
12Filter EG                   modEnv -> filterFc        Partial
133 indep. EGs (amp+flt+pit)  volEnv + ONE modEnv       None
14Two LFOs                    vibLFO + modLFO           Direct
15LFO waveforms               triangle only             Partial
16Tremolo + wah (amp/flt LFO) modLFO -> vol / filter    Partial
17Level / pan                 initialAttenuation / pan  Direct
18Velocity layer switch       velRange per zone         Partial
19Drum key-map + per-drum     zones + keyRange + gens   Direct
20Drum choke / alt-group      exclusiveClass            Direct
21Reverb / chorus SEND level  reverb/chorusEffectsSend  Partial
22Reverb/chorus/variation FX  (send amount only)        None
23Insertion effects           (none)                    None
24Part / master EQ            (none)                    None
25SysEx / NRPN deep editing   (limited modulators)      None

The main wall is SF2 has only two envelopes: volEnv is volume-only, and the single modEnv must be shared between pitch and filter. AWM2 has independent amp + filter + pitch envelopes per element, so you can reproduce a filter EG or a pitch EG, never both at once.

This is why you're better off working with a subtractive synthesis sampler/engine that either takes .sf2 and extends it further or baking in the main wall and accept it's baked in and not going to sound quite the same for most voices. Even the pitch mapping from XG to .sf2 is not so easy either.