Kahenraz wrote on 2022-02-01, 10:41:

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 […]
Show full quote

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?

usb midi is a solved problem, the first ones were not using usb bulk transport, they sucked. everything has been using bulk transport for two decades, MIDI runs at 31 k/bps. USB 1 has 12M/BPS. USB 1 runs less than 12mbps (has like 1m/bps of overhead for messaging) but you aint dropping frames because latency. Thats not talking about highpseed usb or (480M/bps).. Now USB does poll rather than use an interrupt but even usb 1 should way outpace midi that it becomes an issue.

Now your usb recording interface might have loads of latency depending on its buffers + speed its operating at, and if your using asio driver or not, but thats another story!

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

I have no real hardware to test this but on SC-VA (that emulates SC-8820 and on paper also has 64 voice polyphony) it works without noticeable issues.

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.

It does not work this way. When a player plays a Midi file everything becomes a sequence of individual real-time messages that is sent to a Midi out port. Depending on the software stack some components may do additional buffering of message bytes that can also alter the timing of messages received by a synth through its Midi in port.

Well, another thing is that the format in the game is XMI rather than MID and we don't know how this file was adapted.

I'll probably attempt some comparisons.

firage wrote on 2022-02-01, 15:18:

Well, another thing is that the format in the game is XMI rather than MID and we don't know how this file was adapted.

I'll probably attempt some comparisons.

In theory XMI -> MID(SMF) conversion is lossless (apart from loop points that are not standardized is SMF but exist in XMI). I do not think any conversion/converter do such thing as increasing polyphony.
But if someone can extract the original XMI from the game FSMP supports XMI playback so it can be used for comparison/testing.

It was such a massive rabbit hole to set up an old version of EverQuest and record using the AWE64 Gold on my 98SE Tualatin.

This is in the client, EverQuest Trilogy release version from 2001, and an emulated server. SPDIF out and all stock configuration, no external SoundFont.

Happy with this as the baseline in comparisons between players, synths or sound cards.

And just the comparison snippet part:

BloodyCactus wrote on 2022-02-01, 12:57:

Kahenraz wrote on 2022-02-01, 10:41:

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 […]
Show full quote

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?

usb midi is a solved problem, the first ones were not using usb bulk transport, they sucked. everything has been using bulk transport for two decades, MIDI runs at 31 k/bps. USB 1 has 12M/BPS. USB 1 runs less than 12mbps (has like 1m/bps of overhead for messaging) but you aint dropping frames because latency. Thats not talking about highpseed usb or (480M/bps).. Now USB does poll rather than use an interrupt but even usb 1 should way outpace midi that it becomes an issue.

Now your usb recording interface might have loads of latency depending on its buffers + speed its operating at, and if your using asio driver or not, but thats another story!

It isn't necessarily the actual USB to MIDI interface, especially newer ones. I did a bit of testing on my main machine a few years ago and ended up getting a Roland USB MIDI interface and it did work way better than the Chinesium one I had before that.

But the actual USB controller can cause issues depending on the polling rate, if anything else is using the same USB controller (since USB is of course serial) or if something else is causing delays for the USB ports.

Having timing sensitive stuff going through USB is just asking for trouble IMHO.

Here's another in-game recording, using the Audigy Platinum sound card. Has its nice moments, but it's clearly not as polished polyphony wise as the AWE64 rendition.
The 4MB GMGSMT SoundFont was the default with this install disc, if it's making any difference. (I don't think it does.)

Comparison snippet:

The largest and most final of the Creative Labs Soundfont files seems to be the 28MBGM, which came with the X-Fi cards. Give that one a try, as it sounds pretty nice. The 2/4/8MB GM Soundfonts are relatively low-tech and unpolished. The 2/4/8MB GM fonts are identical apart from size/compression, so you should test with the 8MB one, for the sake of quality.

I’ll definitely check out how that SoundFont sounds.

The game loads its original SoundFont when it’s working correctly. I think it has to recognize the card in order to access the synth properly; here it’s definitely not right. The Audigy must be too new. SB Live cards might work more as intended.

RetroGamer4Ever wrote on 2022-02-07, 02:46:

The largest and most final of the Creative Labs Soundfont files seems to be the 28MBGM, which came with the X-Fi cards. Give that one a try, as it sounds pretty nice. The 2/4/8MB GM Soundfonts are relatively low-tech and unpolished. The 2/4/8MB GM fonts are identical apart from size/compression, so you should test with the 8MB one, for the sake of quality.

What are the file names for these soundfonts?

RetroGamer4Ever wrote on 2022-02-07, 02:46:

The 2/4/8MB GM Soundfonts are relatively low-tech and unpolished. The 2/4/8MB GM fonts are identical apart from size/compression, so you should test with the 8MB one, for the sake of quality.

This is definitely not true. The 2/4/8 MB soundfonts from Creative/EMU do not even contain the same number of presets, not to mention samples. Arguably the 4GMGSMT.SF2 version is the most feature rich since it contains the most variation tones and drum sets.
BTW even the 28MBGM.SF2 from the X-Fi CD has rather serious instrument range definition bugs. (e.g. try it with the attached Storm.mid from Duke Nukem3D. The cricket mimicking effect is completely missing at the beginning).
Moreover it misses any GS/MT-32 variation tones. So yes, it has higher quality samples but nowhere 'more polished'.

Has anyone ever attempted to fix the bugs and improve these soundfonts?

Kahenraz wrote on 2022-02-07, 11:36:

Has anyone ever attempted to fix the bugs and improve these soundfonts?

A decade ago I fixed some bugs in the 4MB one since it was the most feature rich and had the best performance/size ratio.

Ahh, and I almost forgot:
28MBGM.sf2 has one of the biggest bugs ever made in an 'officially' released soundfont by Creative/Emu. Namely it completely misses Picked Bass (34) preset from the capital tones. So its usage is absolutely not recommended. Its characteristics is almost the same as the 4/8 MB soundfonts (but has more serious bugs) and there are much better sounding/less buggy soundfonts in the ~30MB range.
Here is a test file attached to confirm the above bug:

So overall considering the size the 4GMGSMT.SF2/CT4MGM.SF2 (exactly the same one) is a decent 4MB SF2 soundfont with 329 (GS/MT-32) presets. Contrary the 28MBGM.sf2 has only 135 presets (missing Picked Bass and no GS/MT-32 presets) and considering its size it is a rather mediocre 28MB soundfont.
The 8MBGMSFX.SF2/CT8MGM.SF2 (exactly the same one) is somewhere between the above 2: it has 137 presets (so no GS/MT-32) but at least it has fully implemented capital tones (Picked Bass is not missing). Otherwise it also has the same characteristics as the others.

On the project1999 message board I posted earlier on post 104, user Maestrobob merged the Everquest soundfont synthuser.sf2 with 1mgm.sf2 from Soundfont Librarian using Creative Vienna.

Can you try that soundfont with the Audigy and see how it compares with the AWE64?

On another subject, I believe I found the original SBK soundfont, Genmidi.zip on the FreeBSD soundfonts server that synthuser.sf2 is based on.

But synthuser.sf2 does have some modifications, at least with some additional instruments.

The Audigy produces the same music in the game regardless of SoundFont or lack thereof, regardless of which synths are selected or enabled in Multimedia settings. Before and after updating the (VxD) drivers, also.

The SB Live! Value II (CT4830) is sounding great, however! Quite similar to the AWE64, with less reverb.

Since this topic helped convince me to buy yet another module (SC-7) I'd like to contribute.

I don't have a Roland 8820, but I do have a D70 which is supposedly the same as an 8820 but with digital audio support, as well as 44 and 48KHz output versus the 32Khz on the 8820/8850. It has the same 64 polyphony limit of the 8820, not the 128 limit of the 8850.

Here's a recording directly from the D70's USB audio, originally capture at 24bit 48Khz.

I believe I hit the out-of-note-polyphony issue with my SC-55 with the TIE Fighter Collector's Edition intro tune. There's a part with four (I think) rapid string runs in quick succession; my Yamaha MU80 and the Sound Canvas VA play all the string run notes, they're all clearly audible, but the SC-55 only the first few per run and the rest are completely missing.

I wasn't sure about the root cause, but I think the lower of polyphony of the SC-55 (v1.21) sounds like a plausible explanation.

Actually, I've done a recording of the MIDI, so I'll to try play only the string channel in solo. If this theory is correct, then all notes should play out with the channel soloed.

This is an interesting thread.

If I've read all this correctly, this particular game was released late enough that it was designed specifically for Creative AWE32 synths, so it actually sounds different\wrong on non-Creative synths?

Are there any other games that are known to have been produced this way? I have heard that Final Fantasy 7's music was written with Yamaha XG in mind. I know that some games have a "Sound Blaster AWE" music option in their setup program too, like Descent, and it will apply it's own specific sound font\bank when to load the game... though that's really the only game that is coming to mind for that option right now.

Aside from these, it seems like nearly every other game was designed for and has the least problems with Roland GM modules and the devices that mimic them.

Is there a dedicated thread or chart anywhere that has compiled information like this, where a specific game's music implementation either only works properly on or sounds different\best with a specific MIDI device? If not, I think there should be. Would be a good place to compile things like what has been found in this thread, as well as, for example, games that benefit greatly from a GUS, CM32 or other devices. We all like to tinker with different MIDI modules and wavetable cards, but sometimes there may objectively be a "best" device for a certain game.

Final Fantasy 7 was designed for a Yamaha XG synthesizer, and even shipped with a Yamaha software synthesizer on the CD. It also provided a custom sound font for the AWE32. Fallback was MIDI without XG, which was probably meant for the Windows built-in Roland Synthesizer. I doubt it was ever even tested on a genuine SC-55. The XG MIDI files actually have XG-specific notes, so this is the optimal option, and how it was meant to be played back.