GEMS, SMPS, and standard Sound Drivers

[kool kitty89]

What exactly was the resoning behind supporting GEMS as the main music and sound engine for Western Genesis/MD dev kits, and indeed using it in a number of first party western dev'd titles?

I seem to recall it being mentioned that GEMS was intended to make the most of the sound resourses on the Genesis, but I'm not entirely sure what that meant.

Even if they did want to promote GEMS as such, why couldn't SPMS have been supported as well, alongside it? It seems to have been the standard Sega engine for any 1st party JP stuff (other than titles predating it), so why not support it in the west as such?
Was it a decision from Japan, or something SoA decided?

[retrospiel]

Early games like Dick Tracy, developed by Sega of America, use SMPS Z80.

Although a lot of you people seem to imply this, Japan had no interest in SOA's games being bad. However, it seems that SOA itself didn't care much what quality their games had - and I am referring to marketing, not the game designers - hence the awesome quality of releases such as Art Alive, Fantasia, Spinball, Jurassic Park, Ariel, etc.
They gave their programmers little to no time to finish their work and even released unfinished betas (as is well documented in case of Fantasia). They had 4 months for Spinball, same for Jurassic Park, even less for Art Alive and Power Rangers on SCD - and just about twice as much for Sonic 2 (!):

In September 1991, four months after Sonic The Hedgehog's release in North America, I'd managed to reunite two of the three key Sonic Team members (Yuji Naka, the Sonic programmer and team leader, and Hirokazu Yasuhara, the designer) at my Sega Technical Institute. They were ready to start work on their next project, and so I asked marketing the obvious question, "would you like another Sonic?" Bizarrely, the response was, and again I kid you not, "no, it's much too soon." So we found another game to make, and in November, as we were getting started, marketing came back and said "oops, we do need that game, and we need it now." So the team lost two months out of an eleven month schedule!

"Dust Hill" was one of dropped levels from original plan of Sonic 2. Sega of America did want to sell the game at Christmas time of that year, I had no choice, but decided to cut some levels out from original plan. So if Sega allowed me to use enough time to pursue the plan, The Dust Hill zone would be in the game.

[kool kitty89]

I never implied SoJ intentionally sabotaged SoA's software houses, indeed I don't think the major internal problems really came about until around 1993, and even then, it wasn't the actions of one side alone, or even influence of Sega alone, which was fully responsible for all those problems. (different portions of both branches were in conflict, and not in a consistent manner either, plus Sony's emergence seems to have had an effect as well, especially on the Saturn's redesign and rushed released, etc)

However, I wouldn't have ruled it out that SoJ wasn't totally enthusiastic about SoJ handling their own software development, or a lack of facilitation for bringing SMPS into the Standard SoA SDKs being distributed to 3rd parites and used by STI.


Apart from that, I'm curious as to just how and why GEMS came into existence. (especially that remark -I think TmEE made- about it being meant to make the most of theing, not wasting anything -that came up after I commented that GEMS sounded a lot like 2-op FM instruments at times, and he corrected be iirc)

[Da_Shocker]

Early games like Dick Tracy, developed by Sega of America, use SMPS Z80.

Although a lot of you people seem to imply this, Japan had no interest in SOA's games being bad. However, it seems that SOA itself didn't care much what quality their games had - and I am referring to marketing, not the game designers - hence the awesome quality of releases such as Art Alive, Fantasia, Spinball, Jurassic Park, Ariel, etc.
They gave their programmers little to no time to finish their work and even released unfinished betas (as is well documented in case of Fantasia). They had 4 months for Spinball, same for Power Rangers on SCD, same for Jurassic Park and probably less for Art Alive - and just about twice as much for Sonic 2 (!):

Jurassic Park was a good game for the Genesis and clearly the best 16-Bit version out. Power Rangers for the SCD was simply garbage and was easily the worst SCD game ever.

[retrospiel]

Whether one likes the game or not, it definitely could have been much better - and it probably would have been if they were given more than a few months.
If you read the interview, you'll be surprised to find out that they had already determined the marketing campaign before Blue Sky was even assigned to work on the game.
SOA didn't know or even care if the game was gonna be good!

- Imagine if Disney launched a phenomenal marketing campaign for a direct-to-video release of "Lion King 5: Timon & Pumba's Lazy Days" - before having even assigned some amateur studio to draw just one single cell.


@kool kitty89: Yeah, I definitely am curious as well as to why GEMS was developed. SMPS seems like the ideal sound driver for all kinds of different music. Why one earth would you think GEMS would be the better choice to use and give out to third parties ?

One explanation might be that SMPS might be less user friendly to people that had no experience with programming a Yamaha FM synthesizer. It could also be possible that SMPS relied on some Japanese workstations that just weren't practical to import. But even if they were using standard computers: This was pre-Windows 95 and there were various competing computer systems in use that were incompatible to each other, and things were even worse from country to country.

[Da_Shocker]

While we are on the subject of sound drivers does this have anything to do with the SF series and MK series. Both SF2SCE and SSF2 had all the voices of the SNES version yet they sounded muffled. MK1 and MK2 had voices that sounded good yet they were always missing tons of voices that were in the SNES version and the cart were always the same size.

[retrospiel]

As the entire sound hardware on the SNES relies on samples which if stored uncompressed would take up a lot of valuable space on your cartridge, it has hardware support for ADPCM compression which allows 16-bit PCM CD quality samples such as voices to be stored at 4-bit (1/4 of the original size) without any loss of quality noticeable.

The same game running at the same video resolution (usually 256x224, which is the common video mode available on both MD/G and SNES and used in Street Fighter II) would use a lot less ROM space on SNES - even if you were to cut down sample quality on the MD/G variant.


Mortal Kombat is a different story. It was developed by different teams as far as I know so that's probably why there's stuff missing.

As for the voices being clearer: Even if the stored sample on your cartridge is of top notch quality thanks to ADPCM, the SNES uses a number of filters and has other limitations that muffle the audio quite a bit (think of the model 2 Genesis).

The Mega Drive's sound hardware is way more bare bones and you can get it to output very clear speech if you want to. Your only limit was the size of the cartridges. And back in the day that was an issue.

[GeckoYamori]

From what I heard, GEMS had some MIDI support which is a tremendous help for composers. With SMPS you still composed in the old way, so musicians essentially had to be programmers as well.

[Silanda]

One explanation might be that SMPS might be less user friendly to people that had no experience with programming a Yamaha FM synthesizer.

This would be my bet. IMHO a lot of American musicians of that time, while talented composers and arrangers, were not programmers or sound designers. There is a parallel in the music industry of that time as well. In the 70's most keyboard players were also good synth programmers, but thanks to the wide adoption of the FM based DX7 in the 80's, along with it being notoriously hard to program, by the 90's there were a lot of new players who just relied upon presets and samples (sampling having became increadibly popular in the late 80's/early 90's).

I believe the trend was carried through to video game music as well, perhaps exacerbated by the increasing amount of pure musicians in the industry where as earlier there tended to be more multi skilled musicians/programmers.

No idea if I'm right, but that's my $0.02.

[TmEE]

From what I know, nearly all other music setups were MML based and reuqired musicians basic programming knowledge... GEMS on the other hand used a sequencer setup and supported realtime music listening where as things like SMPS did not.
The Krisalis one that Matt Furniss used was a nice setup that used Atari ST + tracker and had realtime sound support, but it seems it never got licensed for use with anyone else, which is a bit of a shame since it is some of the best sounding setpus for MD that ever existed.

[Silanda]

It would be interesting to see what the setup Jesper Kyd used was like too, seeing as it was allegedly created because GEMS was a bit crap.

[Chilly Willy]

As the entire sound hardware on the SNES relies on samples which if stored uncompressed would take up a lot of valuable space on your cartridge, it has hardware support for ADPCM compression which allows 16-bit PCM CD quality samples such as voices to be stored at 4-bit (1/4 of the original size) without any loss of quality noticeable.

I wouldn't say without ANY - most people think the SNES sounds muffled, and that's a direct result of the compression. It's not bad, but it's certainly noticeable to most people.

[roundwars]

I wouldn't say without ANY - most people think the SNES sounds muffled, and that's a direct result of the compression. It's not bad, but it's certainly noticeable to most people.

Are you sure it's the compression that causes that? The SNES has a few other limitations that could cause it to sound muffled, such as the 32khz ceiling on the sound chip's output. Though what's probably more relevant is the fact that a lot of samples in SNES games are stored at much lower sample rates than that in order to save space.

[TmEE]

The compression, low sample rate, rather strong low pass filtering on some models, very low quality samples, forced interpolation are all factors in the muffled sound.

[kool kitty89]

While we are on the subject of sound drivers does this have anything to do with the SF series and MK series. Both SF2SCE and SSF2 had all the voices of the SNES version yet they sounded muffled. MK1 and MK2 had voices that sounded good yet they were always missing tons of voices that were in the SNES version and the cart were always the same size.

Really, I find the Genesis versions chrunchier, but the SNES versions more muffled (PCE versions are better than either). The poor quality and limitations of the Genesis PCM playback come from several factors. For one, the samples are uncompressed, thus either take up a lot of space at high quality (so fewer samples), or muct be cut in terms of resolution (bit depth) and/or sample rate.
On top of this there's the Genesis's rather poor mechanism for playing samples, using CPU resourse (usually the Z80) to control sample playback in software and write directly to an 8-bit DAC avaiable on the YM2612's 6th FM channel (disabling the FM on that channel). It requires very tight code to produce good quality playback, though this wouldn't have been anywhere near as problematic if Sega had the interupt lines from either the Z80 or 68k hooked to the YM2612 for precise timing. (really odd they didn't have that, even if if a patch wire was needed for a mistake on early revisions it sould have been there) This has come up before.

The SNES has dedicaded PCM playback hardware and ADPCM decompression capabilities, it relies on manipulation of digital samples entirely for its sound production.

As the entire sound hardware on the SNES relies on samples which if stored uncompressed would take up a lot of valuable space on your cartridge, it has hardware support for ADPCM compression which allows 16-bit PCM CD quality samples such as voices to be stored at 4-bit (1/4 of the original size) without any loss of quality noticeable.

I'm not sure the entire sample is decompressed from asingle stream of 4-bit data, I think it's a bit more complex than that. (like multiple peices of 4 bit data with seperate delta and samples)

Anyway, there will be artifacting from ADPCM, especially at low sample rates, part of the reason for the rather intense filtering and interpolation. (though it's arguable if those did more harm than good -especially being forced -optional, or variable filtering or interpolation would have been very useful, and indeed there are cases where no interpolation sounds better in general -in emulators)

At best, the SNES is capable of 32 kHz, so that combined with the heavy filtering and interpolations plus compression artifacts, makes it fall a good bit short from CD (or MP3) quality audio.

The Mega Drive's sound hardware is way more bare bones and you can get it to output very clear speech if you want to. Your only limit was the size of the cartridges. And back in the day that was an issue.

Size, code quality (for software playback), and CPU resourse used. (you can even do software decompression -Chilly Willy has been experementing with soem fairly hefty Z80 driven compressed audio streams played through the YM2612 over at spritesmind I think)

As for the SNES specifically, tomaitheous mentioned this in another discussion:

The samples are 16bit format per sample (the original ADPCM saturated(or wrapped depending on the hardware decoder) to 10bit and 12bit). The '4bit' value is a delta, accumulating into an index to another delta layer (the actual sample), that itself accumulates. It's a variant of ADPCM in that it adds per block filtering typing and repeat markers. ADPCM is pretty good compression relative to its uncompressed source, then add in filtering and it's even more accurate to its original form. But the 64k ram has nothing to do with streaming audio. You just need to transfer 154bytes per frame (vblank interrupt).

The compression, low sample rate, rather strong low pass filtering on some models, very low quality samples, forced interpolation are all factors in the muffled sound.

It's interpolation mostly I think. Judging by real hardware and emulators set to varous audio settings, the low pass filtering doesn't have that extreme of an effect; it's the interpolation that really does it, soemthimes changing the sound of intruments entirely. (liek in Doom at E1M1, th elead instrument sounds like a cello with interpolation, but more like a guitar with interpolation off)

The main thing noticeable from the compression is soem chrunchy atrifacts, apparently most noticeable on lower sample rate stuff. (a bit like the artifacts int he 32x's PWM in some respects)