Software Synthesizers:
These can be further broken down into general software synths that use sound fonts (FluidSynth and BassMIDI are two prominent examples) and software synths that emulate specific devices or technologies (Munt, SY-YXG50, and SC-VA being most relevant). Obviously FluidSynth and BassMIDI are very flexible, but they tend to be limited by the capabilities of the underlying sound font formats. In comparison, Munt, SY-YXG50, and SC-VA can emulate their target devices (MT-32/CM-32L, MU50 family, and SC-88 respectively) potentially to the limits of the rendering device.

EMU devices and expandable HW synth cards.
High quality software synthesis wasn't always possible. AWE32/64 cards didn't require the CPU muscle of the software synths above but they have lower limits for basic parameters like polyphony and soundfont size. They offered flexibility in sound which would otherwise require a rack of gear or an expensive pro synthesizer like a Roland JV series. Other cards with loadable soundfont capability from Turtle Beach, Terratec, and so forth are essentially in the same boat, although they usually use a hardware MPU-401 interface and not emulation or a dedicated interface like the EMU cards and often used less common soundfont formats.

YMF-PCI XG emulation
I believe this is hardware assisted, but that's less of an issue in these days of high CPU performance. Still, the card is great for other reasons (real OPL3).

Sound modules/sound cards/daughterboards
The SCB-55 is essentially the SC-55mk2 module with the front display and buttons removed, then shrunk down into a wavetable daughterboard. The SCC-1A card is essentially an MPU-401 (intelligent mode) ISA card plus the same guts as in the SCB-55. Yamaha also had modules, daughtercards, and expansion cards for their MU family. Some companies (Korg, for instance) made modules and daughtercards but not expansion cards. A large variety of chipsets weren't used in external modules (much?) but were common in daughterboards and sound cards. As for a discussion of SC-55 vs SC-55mk2 vs MU50 vs MU80 vs everyone else, I'll leave that for wiser souls. Suffice it to say, module families were usually very different and sounded quite distinct, although sound cloning/theft did happen.

Thanks for the reply.

If it were possible to use the exact same samples from a sound module on a software synth, an EMU-type synth, or any number of other devices, what makes them sound different? Does it all come down to the implementation of reverb/chorus effects or is there more to it than that?

Has anyone done any experiments like this?

Ozzuneoj wrote:

Thanks for the reply.

If it were possible to use the exact same samples from a sound module on a software synth, an EMU-type synth, or any number of other devices, what makes them sound different? Does it all come down to the implementation of reverb/chorus effects or is there more to it than that?

Has anyone done any experiments like this?

Electronic components used in general, quality of grounding and other noise reduction, make of D/A converters, output stages, etc.

As for "exact same samples", if the devices tested use different methods of synthesis, it is not possible to have "exact same samples" because some methods of synthesis (e.g. FM) are not sample-based.
Furthermore, you would prefer to test without reverb/chorus enabled, because there are different algorithms in use and different types of effect, plus depending on the implementation the effects are applied before or after mixing the instruments.

Obviously you can't use samples on FM synths. All the devices I'm talking about are sample based.

I guess I'm not getting the point of this thread across very clearly.

Ozzuneoj wrote:

Obviously you can't use samples on FM synths. All the devices I'm talking about are sample based.

I guess I'm not getting the point of this thread across very clearly.

Thing is, it's sort of "all of the above". There are differences in samples, differences in effects, differences in supporting hardware (mixers, filters) etc. - and as soon as anything happens in software drivers and settings also come into play. Most of these variables simply can't be teased apart except in situations where you have the exact same core chipset in different kinds of devices (SCC-1 vs SCB-55 vs SC55 mk2) or the different chipsets on the same supporting logic (i.e. different MIDI daughterboards on a reference sound card).

Maybe I'm too much of a philistine, but I just feed whatever I want into the various devices I have at my disposal and use what sounds best, which in the case of MIDI tends to be MT-32 for MT-32 stuff (duh) and Yamaha MU50XG for GM (although the reason I prefer it over MIDI daughterboards probably has more to do with liking buttons to fiddle around with while playing 😉 )

I prefer software synths. But I make my own so I can get them to sound however I like. Better than the original hardware is easily possible these days, but you will run into compatibility issues with certain games.
I'm working on porting a bunch of my software synth code over to a raspberry pi so I can plug that into a old MS-DOS pc for MIDI synthesis.

Modern emulation can get you 99% the same sound as original hardware, but 99% isn't 100% and there's no such thing as perfect emulation. There will always be a difference, but that's not always a bad thing.
The best overall hardware synth is probably the Roland SC-55, but it's far from perfect. At least it's very high-quality and gets you a very "90's" era sound. Sadly it's also pretty expensive.
Several newer 90's games sound best with a Yamaha XG compatible synth, but by that time most games didn't use MIDI anymore. I would probably just use a software synth for those.

Many 80's and early 90's games only sound right with a Roland MT-32, but that synth is far from GM compatible so it sounds pretty bad for any game that's not designed for it. It does have a really nice "80's" sound to it though.

I recommend reading some of the back posts from Hierophant, Cloudschatze, James-F, and Falcosoft as they're some of the bigger MIDI gurus on the forum. Also, Hierophant's blog is a great resource.
http://nerdlypleasures.blogspot.com/

In the last decade or so, a lot of the original hardware synths that became staples in not only the game music industry, but the real music industry as well, have been porting over to software synths with varying degrees of accuracy.

Take, for example, the Virtual Sound Canvas plugins and soft-synths. Presumably, these have 100% identical ROM sets to the original hardware. They don't sound QUITE perfect, but pretty darn close. The differences come mostly from the accuracy of the playback engine. In the original hardware, it was a combination of ASICs and DSPs and analog circuitry. In the software, it's (obviously) an emulation of those components. Usually it'll be an emulation of the process, and not the components themselves... such that Roland isn't going to the trouble of trying to recreate the reverb's character exactly, but rather to throw a pre-baked reverb algorithm at it, tweak the parameters to match closely enough, and job done. The Korg M-1 soft-synth is a good example of this. It just doesn't quite have the buttery smoothness of the original box, and some of the effects just don't sound right at all.

Sometimes, the synth and filter parameters aren't 100% spot on either, which makes the presets sound just a little bit different. Maybe the balance between samples in a mutli-part preset aren't matched properly, or the lowpass filter has a slightly different knee, or the modulation LFO's range is off by a bit. It's all attention to detail that apparently nobody can justify for a recreation of an old ROMpler that'll sell for $50 to not that many people.

You have to understand, Roland had Eric Persing -- who is a god among men -- sitting there tweaking these sounds for who knows how long to get them just right. When you fiddle with the playback or effects engine, those careful tweaks come undone.

The takeaway there is that getting a patch set today that sounds better than those old ROM sets isn't just a matter of pointing a microphone at a sound source, hitting record, and slapping in a 24-bit, 96kHz sample. It's about the playability, expressiveness, and adaptability of that sample -- how does it fit in context of a preset? Is the attack good for a single hit? For a run of notes? Is it going to sound monotonous when playing it? Does it take so long to develop that it can't be used in staccato? What about loop points? How does the timbre change as the sound plays? Can it be looped without a stark change in tone? Does it need to be looped at all, or do we just throw memory at it? What about the release? Is it natural with just ADSR parameters, or do we need a different sample for that?

So, even though people can now create Sound Fonts of many MB or even GB, without the obsessive skill of a dedicated sound designer, it won't be as universally effective as those old 4MB ROM kits.

Then there are proper synthesizers... Some are meaning to sound like an old Juno or DX-7 or a Moog. Those sounds are defined as much by the analog circuitry as anything else, so just creating waveforms in a computer and applying DSP to them doesn't create the same texture as the original. For that, you have to model chaotic stuff like the interaction of hand-wound inductors and microphonics of ceramic capacitors and the non-linear distortion of a saturated transistor, or a leaky hybrid analog/digital ICs, and who knows what else. Companies like Waves do something similar for analog signal processing gear, but it's not that often that anyone goes to that level of detail for a synthesizer. But we've only (relatively) recently been able to do it, so it's still very much an art and science under development.

Thank you! That's there kind of information I've been trying to dig up, but I hadn't seen it presented in that way. It makes perfect sense that the process of translating MIDI data and a bank of samples into accurate playback is far more complex than it may at first seem. Being this complex, there would be a lot of different ways to design a sample based synth (I'm thinking mainly about the huge variety of wavetable card that all sounds completely different). The quality of all of the basic parts (effects, filters, etc) would greatly be based on how much effort and expense was put into making every part work together, and the quality of the samples is only a small part of it. Would it be accurate to say that the end result quality of the MIDI playback relies on how well a device is tuned for the particular samples being used, and how well those two factors match up to the originals used by the composer of a track?

This would explain why devices (including soft synths) that can use 3rd party sound fonts or samples can have very inconsistent quality... because it hasn't had the personal tweaks done for each sample like a high end Roland or Yamaha synth. And even with a well designed piece of MIDI hardware or software, simplifying it to save costs or to lower hardware requirements will keep it from sounding perfect all the time.

More input on this is welcome. This is a very interesting subject. 😀

Ozzuneoj wrote:

Thank you! That's there kind of information I've been trying to dig up, but I hadn't seen it presented in that way. It makes perfect sense that the process of translating MIDI data and a bank of samples into accurate playback is far more complex than it may at first seem. Being this complex, there would be a lot of different ways to design a sample based synth (I'm thinking mainly about the huge variety of wavetable card that all sounds completely different). The quality of all of the basic parts (effects, filters, etc) would greatly be based on how much effort and expense was put into making every part work together, and the quality of the samples is only a small part of it. Would it be accurate to say that the end result quality of the MIDI playback relies on how well a device is tuned for the particular samples being used, and how well those two factors match up to the originals used by the composer of a track?

This would explain why devices (including soft synths) that can use 3rd party sound fonts or samples can have very inconsistent quality... because it hasn't had the personal tweaks done for each sample like a high end Roland or Yamaha synth. And even with a well designed piece of MIDI hardware or software, simplifying it to save costs or to lower hardware requirements will keep it from sounding perfect all the time.

That's (a rather amusing but) highly idealised viewpoint 😉 🤣 . Even most high-end synths don't have tons of patches that sound "excellent".

Case in point, I picked up that expansion card for my Roland JV-35 that contains all of the patches of the JV-80 and JV-1000 (two respectable synths that are still looked on positively 25+ years on), and raved about how great some of the patches were. Truth to be told, those sounds make up maybe 20-25% of the patches, and most of the "greatness" only came out after tweaking the parameters for most of them.

It's often about what you can do to the patches, i.e. the tweakability of the synth, but it can also be about a characteristic sound, rare set of features etc. When you read synth reviews (and play around with them on your own) you will find various things.
With some synths a particular feature stands out (often the filters) but the presets and other capabilities kinda suck.
Sound Canvasses and the like are generally looked down upon because they are lacking in parameters that can be changed. Some very cheap synths get called "secret weapons" because they look like Sound Canvas-like preset boxes but actually have real synthesizing ability under the hood (but you have to dig through the settings to access that).
In other cases, the good is actually in the bad: the synth has a set of flaws, bugs, or undocumented features that make it possible to get out particular sounds that cannot be obtained with most other synths, despite having mediocre patches and official features at best. In some cases the mediocrity of the output sound makes the synth be able to cut through other sounds and make it stand out in a mix. Some synths have fatal flaws instead (e.g. Korg Prologue bugs, for a recent example).
Other synths are remembered for their particular sound. This is valid for many analogue synths, but also certain digital ones (e.g. Yamaha DX7). Sometimes this is only within a particular music genre, while people in other genres abhor the "noise" produced by the synths involved.
Etc. The sky is the limit.

Remember, synthesizers are musical instruments in the first place, while using them as musical accompaniment boxes for games is merely an alternate use: It's not about reproduction, it's about original creation.
Basically, the only moment a synthesizer is used for the exact reproduction of real instruments is when the composer can't be bothered/has no time/has no funds to ask real musicians to perform for them and/or lacks the capacity to record live acoustic instruments. Then they will be happy to own a electronic instrument to reproduce those sounds for them, even though these days they'll likely use some advanced softsynth of the particular acoustics instruments they need.

Ozzuneoj wrote:

With gigantic (sometimes in the GB size range), super high quality sound fonts and tons of processing power, how are these not "the best" at everything? It seems like someone could easily use sound fonts that match the standard Roland sounds from years ago, possibly with higher fidelity and cleaner outputs from a modern sound card. And yet, this seems like a fairly unpopular option.

I've actually tried going this route and seeking progressively better quality sample libraries/sound modules for playback of General MIDI. What I discovered is that there is a lot more than raw sample quality that is a factor.

I've found that as the raw quality improves, other limitations of General MIDI composition and playback become far more apparent. It's almost like reaching the uncanny valley of General MIDI. Everything just sounds a bit "off" due to the lack of nuance and detail found in a real live performance. These nuances are masked with less accurate/lower quality samples. It is like it's easier for the brain to accept it's listening to something synthesized as opposed to a live instrument.

Modern professional sample libraries get around this by including numerous articulations for instruments, true legato, round-robin samples, and so on. All these elements allow composers to construct digital soundtracks in a manner that more accurately recreates the expected sound of a live performance. Without all those added elements, raw sample quality doesn't go very far.

Yeah, the uncanny valley problem is an issue I'm always dealing with as well with my main saoundfont project.
SF2 doesn't support roundrobin and other advanced synth features so I have to get it to do weird tricks to get similar results, however this means that it looses compatibility with most of the older, more obscure sf2 players.

Even then while this sounds great for some songs, it still sounds really weird for others.

^^ Definitely. It's very hard to make sample-based synths fit perfectly into any composition because there are just so many ways to play a guitar or a flute. Things like vibrato sound better if they're baked in than added via an LFO, but if you're not playing something slow and sweet, you don't want it at all. Listen to any dist. gtr. patch and it usually sounds terrible. The sample may be a perfect reproduction of an overdriven guitar, but guitars are responsive instruments. Samples are rigid.

At least when it comes to reproducing virtualized instruments, we're in a place now where we're technologically between 90s ROMplers and the future -- which I believe is model-based synth. (See Pianoteq for example, or the Yamaha VL70m for an old-school shot at wind-instrument modeling.) With models, you have the opportunity to bring the performance into the algorithm, and that's going to be a game-changer for authenticity. If that's what you're after. (And why I'm still a fan of 90s synths -- they don't sound real, just nice.)

On a different note (no pun intended) the thing that, IMO, made Roland synths so popular is that the sound designers were incredibly good at making the sample flexible. That is an underrated talent. I always thought Korg's stuff sounded FANTASTIC in the demo room. Super huge and swirly and impressive. But that can just be too much when in context of a song, unless it's the central character of that song. Roland's patches often sounded ... nice, if not maybe a little underwhelming, but when you started putting together a song, they just fit in better.

Then you have to think about our common and particular use of MIDI modules around here -- to play game soundtracks. For one, they were often written on a Sound Canvas, so obviously they sound best played back on a Sound Canvas, because the composition was balanced using those patches.

That's analogous to the raw samples vs. playback hardware thing. If the designer had the SC hardware and used samples that worked with that playback engine, that DSP, those filters, etc., then moving it to a different DSP, different filters -- it's not going to sound right unless they're functionally identical.

Likewise, if you took "better" waveforms and replaced the SC samples with those, it's probably not going to sound as good playing back grabbag.mid because now you've disturbed the balance of the instrument. Everything's gotta be re-tuned.