Additional recordings from the Yamaha MU50 and Microsoft GS Synthesizer.

Here is a snippet from the previously attached recordings comparing each synthesizer.

Additional recordings from the Yamaha MU50 and Microsoft GS Synthesizer.

Does anyone here know anything about MIDI files and can help figure out what is happening in the composition that is making all of these synthesizers struggle?

Here is the original MIDI file I used for recording.

I ran the file through two software synths (Soundfont GM & Roland Virtual Sound Canvas) that should handle anything you throw at them and found no discernable difference in polyphony between them through the MIDI player I used, though the GS synth certainly sounded better. That being said, the polyphony readings were not particularly high (mostly in the below 5-14 range, with some peaks into 20s) and I have no basis for comparison, so take that as it is. If those are the correct counts, there should be no reason why any of those synths is missing stuff, unless the instruments in the file simply aren't there in the synth you use to play it back or effects are hampering polyphony. It could be a player issue too, though I couldn't imagine why, as it's not a particularly demanding file.

It sounds like there is some kind of issue with the sustain. I think the notes get played but some synthesizers aren't holding some of the notes properly, to varying degrees.

In the interests of gathering relevant information for the case, I have a few questions.

1. Can I ask about the details of the playback setup please?
-operating system
-MIDI player program
-sound card used
-method of physical connection between soundcard and MIDI devices

2. Also, what is the sub-format of the MIDI file? is it a MIDI file created for GM devices? or a MIDI file created for GS devices? Load the file into FSMP and it will tell you.

3. Third and last question for now 😁 have you ensured each device is hardware reset to default settings for the purposes of these recordings? And also, where relevant - e.g. if other songs have been played previously - that the devices are GM/GS Reset in the software/playback sense, so that no settings from earlier songs are still active?

All of these recordings were done on Windows 10 in stereo with matched settings using a Behringer UCA202 USB ADC and Audacity. I made over 100 different rerecording recently and had a procedure for resetting the MIDI device between each recording using a specially crafted GS reset file which I confirmed to work consistently across all of my Roland synthesizers.

I can guarantee that everything was reset consistently between each recording and I confirmed this by checking the LCD display between resets. I practiced this procedure on all LCD synthesizers before using the same procedure on the non-LCD SC-7 and ST variants.

I spent a couple of days making a lot of recordings (not just this one track) so I had a lot of practice to perfect the workflow.

All recordings were done at 44.1 kHz at 16-bit and this is matched in the ADC capture settings dialog that's buried elsewhere. The recording Hz for the device was matched 100% with what I choose in Audacity.

Because of the consistent output between synthesizers, I can say with confidence that this is not an issue with the UCA202 recording device.

According to this post Make EQ's music (MIDI) sound Significantly Better., EQ came with its own soundfont, and the music was stored in XMI format.

It seems an AWE or Sound Blaster Live would be the best sound card to record the music from, using a player that understands the XMI format.

Kahenraz wrote on 2022-01-30, 01:25:

All of these recordings were done on Windows 10 in stereo with matched settings using a Behringer UCA202 USB ADC and Audacity. I […]
Show full quote

All of these recordings were done on Windows 10 in stereo with matched settings using a Behringer UCA202 USB ADC and Audacity. I made over 100 different rerecording recently and had a procedure for resetting the MIDI device between each recording using a specially crafted GS reset file which I confirmed to work consistently across all of my Roland synthesizers.

I can guarantee that everything was reset consistently between each recording and I confirmed this by checking the LCD display between resets. I practiced this procedure on all LCD synthesizers before using the same procedure on the non-LCD SC-7 and ST variants.

I spent a couple of days making a lot of recordings (not just this one track) so I had a lot of practice to perfect the workflow.

All recordings were done at 44.1 kHz at 16-bit and this is matched in the ADC capture settings dialog that's buried elsewhere. The recording Hz for the device was matched 100% with what I choose in Audacity.

Because of the consistent output between synthesizers, I can say with confidence that this is not an issue with the UCA202 recording device.

Awesome, thanks for the details, you are clearly on top of the workflow. The full answer is beyond my pay grade, but I spent a few minutes exploring some differences in your recordings. I focused upon your 55-121, 55-Mk2, and 55-ST recordings, comparing the first 30 seconds. I found that the marked differences most audible to me, were mainly in the horn section. (yes yes British comedy fans, I hear you)

Image: looping and Solo'ing tiny sections, to isolate sound differences between tracks

My dumb assumption is that these are merely the differences to be expected due minor hardware differences, on top of that era's MIDI standards flux. For example some devices are GS, some aren't. Some MIDI files contain GS instruments, maybe even just for the occasional note here or there. Some don't. Even specific issues like CTF (a minor vagary between different revisions of SC-55) have been the basis of many a detailed discussion on Vogons. Merely highlighting the real world differences between "compatible" synths, once you dig down into the details. But I am not expert enough to say for sure.

If someone can isolate out the significant instruments, I would be happy to do some additional recording.

I don't know anything about editing MIDI files. Is there anyone here who does or know someone or another enthusiast forum that can assist with this research?

Yes, some game composers did use Roland Sound Canvas modules to do their work in GM, with GS embellishments. That's why many titles and GM game music files sound best on GS, while being slightly off on XG, and "normal" on GM synths of good quality.

Kahenraz wrote on 2022-01-30, 02:33:

If someone can isolate out the significant instruments, I would be happy to do some additional recording.

I don't know anything about editing MIDI files. Is there anyone here who does or know someone or another enthusiast forum that can assist with this research?

There are MIDI experts here such as Falcosoft.

I don't mind digging into it when I get time. Just gotta crack the MIDI open and inspect whatever part you're curious about. Why do certain notes not play, or sound a certain way? We only need to look at the MIDI events at the exact point of interest - and correlate those with each particular device's characteristics - to find out.

FSMP (written by the abovementioned Falcosoft) and Reaper (shareware I think) and two freely available programs which are able to read MIDI files and show you the fine detail within. FSMP has a detailed Event view, and Reaper can show graphical representations of the note roll.

This was surprising because of what I had read about synthesizer equivalence. […]
Show full quote

This was surprising because of what I had read about synthesizer equivalence.

My synthesizers tested with assumed equivalence:

Roland SC-55 (v1.21) - Baseline
Roland SC-55mkII (v1.00) - More polyphony, more instruments
Roland SC-88VL (v1.04) - More polyphony, more instruments
Roland SC-7 - Similar to SC-55mkII?
Roland SC-55ST - Equivalent to SC-55mkII but not LCD?
Yamaha MU50 (v1.06)
Microsoft GS Synthesizer (Roland)

In terms of equivalence, I'm not sure what you read, but aside from all being GM compatible, several of these are fundamentally different. Was your assumption that all of the above should be the same as far as SC-55 sound in gaming?
If so:
- SC-88VL is not SC-55 equivalent, only compatible. Hence why you hear different sounds for instruments such as French Horn. Although, it appears to be in 88 mode, not 55 mode, which would yield closer results (although since you have a 55, not sure why that would add anything).
- The rest of the Rolands should be very similar sounding for GM, polyphony and minor quality differences of DAC's aside.
- The MU50 is obviously a Yamaha XG synth, and again, compatible but certainly not equivalent, although I'm going to assume you didn't think it was.
- The Microsoft GS Synth is the same samples in theory at least, but doesn't utilise any effects (except the WinAmp version of it, which allows them), so doesn't really sound like a SC-55.

As for EverQuest - I opened it up and there's no obvious reason why synths should operate differently, there's no MSB or LSB instructions for GS devices, for example. There's a "Balance" MIDI CC being sent, which I've not run into before, but apparently controls stereo balance of stereo patches (probably not related to the issues).
There is lots of reverb and chorus, but that shouldn't be an issue either.
My hunch is it's related to the reset you're using, or maybe polyphony, although if it was polyphony the SC-55ST wouldn't be having the issues and the 55 would.

Question - when you did the recordings, did you - or could you - record one of the synths (that "worked") immediately after a reset and then do the exact same procedure with one of the synths (that "didn't work")?
The two I would limit your scope to would be the SC-55 and SC-55ST, which again should be the same other than minor quality differences/polyphony. Those should reproduce a 95-99% identical result. Right now there's too many synths and too many variables. By reducing the variables we can find out what's actually going on, or give it our best shot.
While I agree like the others that your workflow sounds efficient, since you're not an expert on the MIDI side of things, unless you record the file identically after sending identical resets followed by identical SysEx's, there can always be differences (especially with the ST since you can't "see" what's going on).
It's possible, for example, that some of the synths simply have not been reset - you sent the SysEx to the wrong MIDI port, or the SysEx you use doesn't reset them fully, or you forgot to reset one, and so on. I'm not trying to put you down, mind you - I'm just trying to figure it out. The conclusion is more likely, after all, to be "some minor mistake by Kahenraz" than "half his synths are broken or randomly acting strange".
I can say having been doing similar activities for 20 years, the odds of making a minor mistake are very high, particularly when using multiple synths. It's incredibly easy to forget a step here and there.
I've done it so many times over the years I couldn't tell you how many. 😀

I'll try in the coming weeks to record my SC-55mkII playing the file after a hard reset, and then after using your reset if you upload/link to it, and see if the two recordings are the same as each other, and if either is the same as your recording (minor recording quality differences aside).
On that note - do you have a copy of or link to the GS reset you're using?

My dumb assumption is that these are merely the differences to be expected due minor hardware differences, on top of that era's MIDI standards flux. For example some devices are GS, some aren't. Some MIDI files contain GS instruments, maybe even just for the occasional note here or there. Some don't. Even specific issues like CTF (a minor vagary between different revisions of SC-55) have been the basis of many a detailed discussion on Vogons. Merely highlighting the real world differences between "compatible" synths, once you dig down into the details. But I am not expert enough to say for sure.

Reasonable assumptions but they don't apply here, I'm afraid. Whole instruments missing isn't a "minor hardware difference", especially between two near-identical Roland synths.
There's no GS instruments in this file except for the drumkit, which should still work on all the devices he mentioned. And CTF isn't relevant here, either.
The reasonable expectation is the 4 Roland devices he has, other than the 5th (88VL), should sound extremely similar.

From listening to the links, the SC-55ST track does sound incorrect, but it sounds exactly the same as others, EXCEPT for what sounds like polyphony reduced cutoffs - maybe some lingering gremlins (aka old data). I'd do a full reset and then see if that resolves it.

Relistening, it sounds like all the 55 devices are potentially having polyphony issues, I'm not sure because I don't know what the piece is "supposed" to sound like. However, it's more obvious in a few devices, which is interesting.

I just looked in the MIDI, and from what I can tell, for example at about 1:16 it needs potentially over 28-30 note polyphony before it can playback everything. So all Sound Canvases 55s will be short.
HOWEVER, partial reserve settings (dependent on your individual devices and what you've sent to them, likely default) dictate that MIDI parts 11-16 get the lowest priority for reserve notes. So devices set that way will lose voices, some of which are on the trumpet channels, which appear to be consistently the cut off channels.
This doesn't help with some basic questions, such as, "Why is the SC-55 doing better than the 55ST?", but at least we know generally speaking that the MIDI is borderline unusable on the 55, and that the SC-88 has enough voices.

The likely scenario is that not all of your devices are in an equivalent state. A good test would be to reset the SC-55mkII by "HOLD BOTH INSTRUMENT BUTTONS AND POWER ON. THEN PRESS ALL." and playing the MIDI directly after, recording that, and then afterwards just sending the GS reset and then playing and recording that. If the two states aren't the same, that tells us something. If they are the same, and both have polyphony issues, that also tells us something.

They all sound bad, the composition just doesn't have good GM compatibility. It's probably intended for AWE64 & Live! and sounds the best like that.

https://www.youtube.com/watch?v=dyKIuB_bZKI

"the composition just doesn't have good GM compatibility"
Appears to be a GM file though? I think it's more of what the OP said/I figured iyt, polyphony limitations mean using older synths drop notes. I thought it sounded fine on the 55 myself (polyphony aside).

That link you provided sounds like a higher end Sound Canvas, BTW? (or a remastered recording of a SC?)

Decent GM compatibility in composition would certainly mean it didn't sound like crap on all three of SC-55, SC-88 and MU-50. 😀

No high-end MIDI synth, just the Creative hardware they composed it for. AWE32 magic: https://www.youtube.com/watch?v=pwh7C4OzA5c

Interesting, thanks for educating me. Composed for an AWE32 seems unlikely in 1999, though? (see below for answer)
I grew up with an AWE32, and was constantly frustrated at how bad it sounded, especially compared to the SC-55! This is better than I ever heard it sound, although I'm not crazy about the effects/EAX processing/whatever it is.

Very curious to know what soundfont it's using. I'm fairly sure it's not the default 512KB or so sample set!

EDIT: From the web, it's AWE32 and a custom soundfont: "The music for EQ started coming together as early as 1995 and was composed with the AWE32 soundcard in mind. By the late 90s most games moved away from MIDI soundtracks but EQ still supported ONLY the AWE32 and Sound Blaster Live! cards for hardware sound. From what I can tell, this is because the game client loaded a special EQ soundfont into the card which loaded instruments and drums unique to the game in addition to calling upon instruments native to the cards EMU8000 midi synth chip."

Ironically, these two were the first cards I owned. Wish I'd heard of EQ back then!

Decent GM compatibility in composition would certainly mean it didn't sound like crap on all three of SC-55, SC-88 and MU-50.

I thought it just dropped a few notes and sounded pretty good.

I'm curious to hear an 8820/50 now with these tracks, incidentally.