Here are three recording excerpts from the same track on my SC-55mkII (v1.00) with a simple GS reset. The second of three recordings sounds slightly off but the other two sound identical. There may be other variations elsewhere but I looked only at this part as it seems to be a common place for this issue to occur.

MIDI playback was performed by Midi Player 6.0 (32-bit).

Here is the same excerpt but recorded with the following SysEx. MIDI playback was performed by Midi Player 6.0 (32-bit).

1F0 41 10 42 12 40 00 7F 00 41 F7
2F0 41 10 42 12 40 01 10 02 02 02 02 02 02 02 02 02 02 02 02 02 02 00 00 13 F7

The first two recordings seem identical but the third is different. These recordings were not cherry picked. I made only three recordings each and this is a 33% odds that something will sound wrong, as observed at this specific point in the track.

Kahenraz wrote on 2022-01-31, 09:16:

Here is the same excerpt but recorded with the following SysEx. MIDI playback was performed by Midi Player 6.0 (32-bit). […]
Show full quote

Here is the same excerpt but recorded with the following SysEx. MIDI playback was performed by Midi Player 6.0 (32-bit).

Copy code to clipboard

1F0 41 10 42 12 40 00 7F 00 41 F7
2F0 41 10 42 12 40 01 10 02 02 02 02 02 02 02 02 02 02 02 02 02 02 00 00 13 F7

The first two recordings seem identical but the third is different. These recordings were not cherry picked. I made only three recordings each and this is a 33% odds that something will sound wrong, as observed at this specific point in the track.

sc55mk2_100_af40865_everquest_theme_01_sysex_part.mp3
sc55mk2_100_af40865_everquest_theme_02_sysex_part.mp3
sc55mk2_100_af40865_everquest_theme_03_sysex_part.mp3

Please, try the above referenced 6.1 test version of Midi Player instead since it contains some improvements regarding Midi timing precision.

Here are the full track recordings used for the previous excepts from my SC-55mkII with the special SysEx if anyone would like to see if they can spot anything different as a result. I'm more concerned with the fact that playback can sound so random but this may just be a bad pairing with Roland synthesizers prior to the SC-88.

Falcosoft wrote on 2022-01-31, 09:29:

Please, try the above referenced 6.1 test version of Midi Player instead since it contains some improvements regarding Midi timing precision.

Sure. I had already downloaded the 6.0 stable release and was recording before you had added a direct link for 6.1.

I'll also use the special MIDI file you provided. For the previous recordings I ran my gs-reset.mid and used Midi Player to send the custom SysEx message between each recording.

If I am playing the same track, is it best practice to do a reset before each subsequent playback? This makes the most sense to me but I don't know if this is correct.

Kahenraz wrote on 2022-01-31, 09:30:

If I am playing the same track, is it best practice to do a reset before each subsequent playback? This makes the most sense to me but I don't know if this is correct.

If the tracks themselves contain the reset SysEx (like the attached 2 modified Midi files) then sending resets manually is not necessary (but it does not hurt either). If the tracks do not contain embedded SyEx resets (like your original Midi file) then yes, you should send resets between subsequent playbacks.

Kahenraz wrote on 2022-01-31, 09:29:

Here are the full track recordings used for the previous excepts from my SC-55mkII with the special SysEx if anyone would like t […]
Show full quote

Here are the full track recordings used for the previous excepts from my SC-55mkII with the special SysEx if anyone would like to see if they can spot anything different as a result. I'm more concerned with the fact that playback can sound so random but this may just be a bad pairing with Roland synthesizers prior to the SC-88.

sc55mk2_100_af40865_everquest_theme_01_sysex.mp3
sc55mk2_100_af40865_everquest_theme_02_sysex.mp3
sc55mk2_100_af40865_everquest_theme_03_sysex.mp3

BTW, as a confirmation test if the optimized SysEx works for you at all:
Download the attached Warcraft 2 Midi file and play it after a simple GS reset. You should hear that the bell sounds at the beginning of the track are clearly cut this way:
https://www.youtube.com/watch?v=0ndCckv-lww
After this send a GS reset and then send the optimized voice reserve SxsEx. Now you should hear that the bell sound is not cut anymore.
download/file.php?id=25153

Falcosoft wrote on 2022-01-31, 10:01:

BTW, as a confirmation test if the optimized SysEx works for you at all: Download the attached Warcraft 2 Midi file and play it […]
Show full quote

BTW, as a confirmation test if the optimized SysEx works for you at all:
Download the attached Warcraft 2 Midi file and play it after a simple GS reset. You should hear that the bell sounds at the beginning of the track are clearly cut this way:
https://www.youtube.com/watch?v=0ndCckv-lww
After this send a GS reset and then send the optimized voice reserve SxsEx. Now you should hear that the bell sound is not cut anymore.
download/file.php?id=25153

There is a profound improvement to the bells with this SysEx command. It definitely works.

I'd like to share some comparison recordings for this Warcraft 2 track, but I will do this in another thread as I don't want to derail this one. Suffice to say, it does work on my SC-55mkII.

I've made some new recordings as requested. These were done with Midi Player 6.1 (64 bit) as confirmed by the About dialog. I used the provided everquest_theme_GS_mod.mid which integrates a GS Reset and custom SysEx optimizations. Recordings were made using my Roland Sound Canvas SC-55mkII (v1.00).

A close inspection of the same offending sequence demonstrated that the issue of instruments going missing at random still exists. This is most noticeable with horns. The problem may not be related to horns specifically, but they are prominent in this composition and stand out more than the other instruments.

Here are five recordings with the same sequence I recorded previously. The horns which should appear in the middle of the sequence were absent for the first three but appeared in the last two. That averages as a 40% chance for them to appear, as demonstrated by my limited sample set.

Here is another example where a horn does not appear. There should be a horn at the start of this excerpt.

Whatever the problem is with this track and the SC-55mkII, it's not solved with this SysEx.

Here are all five recordings I referenced in their completeness if anyone would like to inspect them.

Here are the locations of the two sequences I used previously. I don't know how to provide an exact timestamp so I took a screenshot of my loop selection within Audacity.

that the issue of instruments going missing at random still exists. This is most noticeable with horns.

This is polyphony related. They're not going missing "at random", we know they're going missing due to polyphony droputs. At times, the SC-55 has half of the polyphony required, as estimated earlier.

Depending on how partial reserve is set up, notes will drop out differently, as mentioned by Falcosoft.

The French Horn goes missing most frequently as it is a 2-voice instrument (SC-55 patches use 1 or 2 voices PER NOTE). The SC-55 has 24-28 *voice* polyphony, not note polyphony, thus this is reduced to 12-14 *notes* when using 2 voice patches, for an overall note polyphony of 12-14/24-28, depending on the hardware and the mix of 1/2 voice patches used.

The overall finding of this thread can be summarised thusly: "Playing MIDI files on synths with insufficient polyphony for them leads to somewhat random dropouts." Although this can seem weird, as in the EverQuest case, devices like the SC-55/variants are made to handle a few notes over and still sound convincing, not 30 notes over. 😀 So you end up with different results depending what notes the synth cuts and what goes through. As Falcosoft said, MIDI timing can cause variations in this, as does device unique firmware and settings. This used to drive me mad with the MT-32 before I knew what caused it!

What would be interesting is to hear some known Sierra/game tracks with more usual polyphony issues, that were composed FOR the SC-55 (not for a soundfont with unlimited polyphony). I think one of the Castle themes in QFG4 (the one with Space Voice) and the Bookshop track in GK1 are the two Sierra ones I know of. Falcosoft probably knows of good ones like the one he posted.
And then, see if the variation still appears between devices. I believe a mkII solves all the known Sierra examples, for instance.
GK1 track attached here, you can see (hear?) that like the EQ track, it has a lot of sustained notes.
The issues are present after about 30 seconds, in the "main" part of the track.

Spikey wrote on 2022-01-31, 11:58:

The overall finding of this thread can be summarised thusly: "Playing MIDI files on synths with insufficient polyphony for them leads to somewhat random dropouts."

Well, it was a very interesting rabbit hole for me to investigate. I'm happy to have put the time and effort into this for myself and others to learn from. There are lots of great example audio snippets for illustration as well.

This is why the X2GS is an essential piece of hardware to buy for retro gaming. It can go toe-to-toe with the old Roland SoundCanvas/StudioCanvas hardware for game use and those of us who don't have the soundcard/external MIDI module option can just use the newly remade VST MIDI driver with modern SoundCanvas soft-synths, which eclipsed their modules from the get-go. What exciting times we live in!

RetroGamer4Ever wrote on 2022-01-31, 13:28:

This is why the X2GS is an essential piece of hardware to buy for retro gaming.

I hadn't considered this. According to the product description:

Dream 5000 series synth chip with 81 voices polyphony

That's significantly better than even the SC-88. Very impressive. 😀

Kahenraz wrote on 2022-01-31, 13:39:

I hadn't considered this. According to the product description: […]
Show full quote

RetroGamer4Ever wrote on 2022-01-31, 13:28:

This is why the X2GS is an essential piece of hardware to buy for retro gaming.

I hadn't considered this. According to the product description:

Dream 5000 series synth chip with 81 voices polyphony

That's significantly better than even the SC-88. Very impressive. 😀

With the Dream chips, you can make a SoundCanvas/StudioCanvas replacement that has almost all of the GS instruments, several times the polyphony (WITH FULL EFFECTS!), and superior audio quality, with digital and analog outputs that connect to pretty much anything you'd have today. Dreamblaster is working on more advanced models that fit that niche, but his ideas are still in preliminary design phases.

Falcosoft wrote on 2022-01-30, 18:30:

With a modern SF2 soft synth this Midi file easily produces 64+ voice polyphony that is more than double what early SC devices can produce.

Can you confirm whether this will play correctly on an SC-88 or does this still not have enough polyphony?

There is one other thing to note. You are going to have more timing issues with a USB midi adapter than if you use something that isn't USB.

When the USB midi adapters first came out, the latency was really really bad and so was the timing.

There were all sorts of workarounds from upping the USB polling rate to PCI latency settings to core locking settings in Windows.

I still don't trust USB midi adapters to have good latency or timing when compared to a MIDI adapter on pretty much any other interface.

Polyphony issues aside, timing issues could just be because of USB timing and latency issues.

I wasn't aware of this. I'm using a very cheap unbranded USB interface for my recording because it's a convenient way to record from a laptop.

What other solutions would you recommend besides a PCI/ISA sound card with an MPU-401 interface?

I can imagine that latency for realtime messages can be important but MIDI-out only has to send data not respond to it. So if there is any delay then I would expect it to just time shift the whole recording by a few milliseconds, which is fine.

Is this incorrect?