Trevor0402's SC-55 soundfont

[Doomag]

4 minutes ago, TheUltimateDoomer666 said:

No.  SoundFont is a file format for storing samples (which can be anything from sounds ripped from ROMs to recordings of laughter), while a ROM is the data from a read-only memory chip.

 

There is unfortunately no true SC-55 emulator available.

Thanks for clearing that up. I've had one guy complain about this but he complains about everything.

Edited March 4, 2021 by Doomag

[Doomag]

I would like why the gain on this sound font is so low compared to the others. Does anyone know why?

[HackNeyed]

This is amazing! It sounds fantastic and I'm enjoying this so much! My gratitude to all those who made this possible.

 

Hallelujah!

 

19 hours ago, Doomag said:

I would like why the gain on this sound font is so low compared to the others. Does anyone know why?

 

I'm no audio expert but I assume the volume is reduced to prevent clipping and allow some wiggle room for balancing. I've only tested a few songs but in PrBoom-Plus you can change "mus_fluidsynth_gain 50" to 100 or 150 to boost it without (noticeable?) clipping. RottingZombie said there is a similar gain setting in GZDoom they set to 1.

 

Opening it in Polyphone (free soundfont editor) shows that every instrument preset has a custom attenuation dB (like gain or volume) reduction tweak. Because I don't want to change more settings in my PrBoom-Plus config every time I swap soundfonts I've edited a copy of mine to increase the dB of every preset by 6 and it now roughly matches scc1t2.sf2 in loudness. Although I have no idea if there will be clipping issues or if I've thrown off balancing.

SC-55.SoundFont.v1.2b (6 GAIN).zip

[Orchid87]

What is preferrable for this soundfont, fluidsynth or virtualmidisynth?

[OpenRift]

11 hours ago, Orchid87 said:

What is preferrable for this soundfont, fluidsynth or virtualmidisynth?

VirtualMIDISynth

[NightFright]

I have listened to MANY SC-55 soundfonts throughout the years. I own a unit by myself and know what it's supposed to sound like. Sadly, this new soundfont doesn't quite get there. Just compare SC-55 recordings of Doom or Descent with MIDI renderings using this sf2. You WILL notice it's not very similar. Too many instruments sound off, volume isn't right, stuff like that. I am running any new sf2 through the highlights of my favorite soundtracks (Doom, Doom 2, Descent, ROTT, Dark Forces) and can tell quickly when something doesn't convince me.

 

Personally, I am currently using Bill90's Masquerade55. It's probably not perfect in all aspects, either, but using it will give you at least a general idea in which direction it should be going. TheUltimateDoomer666 gave a few excellent hints earlier on in this thread regarding improvements. I can only recommend following his advice if you want to get this to the next level. My suspicion is that 10 MB won't be enough to make this soundfont compelling, even though small filesizes are ofc attractive and resource-friendly even on modern systems.

 

So yeah, I wouldn't call this the "most accurate" SC-55 soundfont, at least not as of now. I do appreciate the intention and determination to create one and hope you won't give up on this, though.

Edited April 7, 2021 by NightFright

[OpenRift]

7 hours ago, NightFright said:

I have listened to MANY SC-55 soundfonts throughout the years. I own a unit by myself and know what it's supposed to sound like. Sadly, this new soundfont doesn't quite get there. Just compare SC-55 recordings of Doom or Descent with MIDI renderings using this sf2. You WILL notice it's not very similar. Too many instruments sound off, volume isn't right, stuff like that. I am running any new sf2 through the highlights of my favorite soundtracks (Doom, Doom 2, Descent, ROTT, Dark Forces) and can tell quickly when something doesn't convince me.

  

Personally, I am currently using Bill90's Masquerade55. It's probably not perfect in all aspects, either, but using it will give you at least a general idea in which direction it should be going. TheUltimateDoomer666 gave a few excellent hints earlier on in this thread regarding improvements. I can only recommend following his advice if you want to get this to the next level. My suspicion is that 10 MB won't be enough to make this soundfont compelling, even though small filesizes are ofc attractive and resource-friendly even on modern systems.

  

So yeah, I wouldn't call this the "most accurate" SC-55 soundfont, at least not as of now. I do appreciate the intention and determination to create one and hope you won't give up on this, though.

Well Trevor (the guy who made this) has been working on making some updates to the font, and is waiting until he has it mostly finished before releasing the next version. The reason I call it the "most accurate" soundfont is because the samples used are taken from an actual ROM dump of an SC-55 Mk II. You can't really say that for most other SC-55 soundfonts because the ROM dump is a more recent development.

[NightFright]

It's definitely the right approach working with source material directly, I won't argue with that. However, this raises certain expectations regarding faithfulness to the original SC55 sound which, sadly, are not met with this soundfont. Yet. 

 

I am afraid I won't be much help with precise improvement hints like "shift instrument x up by two octaves" or "increase volume of drumset y by z%". My musical knowledge isn't sufficient for that. Anyway, I can still assist by telling whether certain key songs of Doom/Dark Forces/ROTT/Descent sound right/better after future updates.

Edited April 7, 2021 by NightFright

[maxmanium]

Question: why is it apparently so hard to get a soundfont that is accurate? I've seen people say that but don't understand why.

[OpenRift]

Just now, maxmanium said:

Question: why is it apparently so hard to get a soundfont that is accurate? I've seen people say that but don't understand why.

Because of the nuances of how the Sound Canvas works. Right now the main issues with the font in its current state is that certain instruments have some pitch issues in certain conditions, namely certain synths. 

[Traveler of the Bizarre]

Hello all! New user that took the plunge due to this thread being so interesting to me (midis are a strange beast that fascinates me).

 

I understand that this soundfont was created by extracting and decrypting SC-55 ROMs.

My question is if the sample quality found in the soundfont is simply the quality that the SC-55 had, or if the samples are being lowered in quality to be smaller so they are more usable on various setups?

I ask this because after downloading HackNeyed's versions of the soundfont (nice job by the way!) I decided to look at the soundfonts themselves using the Polyphone tool that they mentioned.

 

Looking at the sample rates (using the version from HackNeyed), only 12 of the samples have a rate of 44100Hz while the rest are 32000Hz. The other two SC-55 soundfonts I have are from Patch93 and EmperorGrieferus. Patch93's soundfont is MOSTLY 44100Hz, though it dips near the end with many 22050Hz rates and even a 3753Hz rate at the very bottom. On the other hand, EmperorGrieferus' soundfont (which I personally like a lot) has a solid 44100Hz across the board.

 

Obviously simply having higher Hz ≠ more quality, as something can be recorded well and still be terrible to listen to or, as one would judge it in this case, not be accurate to the intent it was designed for. Thus, I'm curious if the samples in the current version of the soundfont are simply the quality that the machines truly had and that the latter two soundfonts simply "made up" higher quality ones as it were, or if indeed the samples are in some way lower quality than they normally would be due to some process of dumping them or the like.

 

Hope that's all understandable, and sorry in advance if this question was answered properly somewhere else on the forum and I didn't notice.

Edited April 21, 2021 by Traveler of the Bizarre

[TheUltimateDoomer666]

@Traveler of the Bizarre
The output resolutions of the major SC-55 devices are as follows:

• SC-55: 16-bit, 32,000 Hz
• SC-55mkII: 18-bit, 32,000 Hz
• SC-55ST: 16-bit, 32,000 Hz
• SCC-1: 16-bit, 44,100 Hz

 

Notice how the sound card version of the SC-55, the SCC-1, actually has a higher output frequency than the module versions.  As far as I know, there are no SoundFonts that use samples ripped directly from the SCC-1's ROM, so the latter two SoundFonts you mention have likely "made up" 44,100 Hz samples via manual recording from an SC-55 module (which results in a waste of space as the source is only 32,000 Hz).

 

All these bit depth and sample rate differences mean there is no one way to make an "accurate" SC-55 SoundFont.  This post says that Bobby Prince used an SC-55mkI (not mkII) to write the DOOM soundtrack.  In that case, an SC-55 SoundFont for DOOM should ideally be replicating the 16-bit, 32,000 Hz quality of the original SC-55.  However, such a SoundFont would theoretically be inferior to one based on the SC-55mkII, or one based on the SCC-1.

 

So, when making an SC-55 SoundFont, do you go for authenticity (for use with DOOM and other games) or higher quality (for general usage)?

 

 

On 4/8/2021 at 7:58 AM, maxmanium said:

Question: why is it apparently so hard to get a soundfont that is accurate? I've seen people say that but don't understand why.

Because SoundFonts are simply collections of samples that require an external MIDI synthesizer (e.g., CoolSoft VirtualMIDISynth).  The SC-55 on the other hand is a collection of samples (the ROM) and synthesizer (the actual hardware).  So not only do you need to extract the module's samples, you also need to emulate the way the synthesizer itself works.  On top of that, the SoundFont format itself has limitations that make it impossible to store some of the SC-55's sample data.  For example, the SoundFont format does not have enough envelope periods.  There is more detailed information here and here.  The SoundFont format is kind of outdated in that regard, and although it is a popular format, its limitations and missing features are one reason why SoundFonts are inferior to hardware MIDI synthesizers.

 

Edited April 22, 2021 by TheUltimateDoomer666
Wording and additional information.

[Traveler of the Bizarre]

@TheUltimateDoomer666

Thanks for the info! I figured that was the case, just couldn't find any easy links to such information. The tidbit about the MkI vs MkII things and accuracy vs quality is an interesting argument. Personally I'm the type who does care a lot about accuracy, but also wants the best quality possible. If one must be sacrificed, I personally go for losing accuracy for quality, at least in the case of music and so long as not TOO much of the accuracy is lost (which I know is vague but it kinda case by case basis imo).

Anyway, very informative post the explains it in layman terms, great job!

Edited April 22, 2021 by Traveler of the Bizarre

[TasAcri]

Unfortunately this soundfont doesn't sound very good.

 

I tested Doom 2 and Duke Nukem 3D.

 

The volume is very low and Duke Nukem some instruments are either too low volume to be heard or missing completely, i can't tell.

Edited June 1, 2021 by TasAcri

[TheUltimateDoomer666]

2 hours ago, TasAcri said:

The volume is very low

This is normal. Sound Canvas VA's volume is also extremely low, even when the synthesizer's master level is set to 127. The reason for this is that if a SoundFont or synthesizer's volume is too high, then louder MIDIs can cause clipping.

Edited June 1, 2021 by TheUltimateDoomer666

[Gez]

On mercredi 7 avril 2021 at 11:58 PM, maxmanium said:

Question: why is it apparently so hard to get a soundfont that is accurate? I've seen people say that but don't understand why.

Because there are more than one factor to get right.

 

MIDI, as you know, is the computer equivalent of sheet music; it tells which note to play with which parameters, but that's all. Then you have the synthesizer. This is the musician. It reads the sheet music and plays the instruments. Which instruments? Well, that's the soundfont (or equivalent, depending on synth). It's a collection of data about how the sound should be. But this data, itself, is made of several parts.

A soundfont's core are its samples. Those are short recordings of, well, actual instruments usually. Then there are a bunch of parameters about how loud it should be, how it behaves to attack sustain decay and other musical mumbo jumbo, and what effects can/should be applied to it to change its pitch and stuff. As discovered by the not-as-convincing-as-hoped soundfont that was based on the actual samples ripped from an SC-55's ROM, getting those parameters right is necessary to properly emulate the original SC-55.

And there's another problem here, in that the guys on VOGONS who ripped those samples to begin with have discovered that some of the effects that need to be applied to those SC-55 samples are not necessarily possible with FluidSynth or supported by the SF2 format. Because the SF2 format was designed to suit the needs of SoundBlaster cards, and not those of Roland cards... I think one of the effects I remember being mentioned was mirroring, where the samples is first played normally and then played in reverse, which when appropriate to use, allowed to only store half of the sample, cutting ROM size requirement for this sample by half. (Something to know about computing in the nineties: every single byte counted!)

 

So there's a lot of work to do to really understand how the SC-55 worked. Getting its raw samples right out of its ROM is just the first step.

[TheUltimateDoomer666]

On the topic of SC-55 accuracy, I noticed that the OP says this SoundFont was created from the SC-55mkII samples, not the original SC-55 samples, which means some instruments may not actually be authentic to what the composers intended.

 

Hexen's music is a good example of instrument differences between Sound Canvas models becoming apparent, because that game features both MIDIs and CD audio tracks of the MIDIs recorded on a "Roland Sound Canvas" (as stated in the game's readme).

 

To elaborate, "Winnowing Hall" (winnowr) has a choir at the beginning. When using this SC-55 SoundFont, the choir sounds pretty human-like (it's a voice going "aah"). This is in line with how the real SC-55mkII's choir sounds, going by this video. But that is not how the choir sounds in the CD audio version of the song (track 13). While the CD audio versions of the songs are mixed a little differently from the MIDI files, it's obvious that the choir instrument in the CD track is a completely different sample set. The choir is much more ghostly and synth-like than the SC-55mkII one.

 

Edit: There are percussion differences as well. In the track "Level" (MIDI: levelr, CD track: 21), a vibra-slap plays at about 10 seconds into the song. The SoundFont sample sounds nothing like the CD track one.

 

To avoid confusion, perhaps it should be specified in the SoundFont's title that it is an SC-55mkII SoundFont and not a SoundFont based on the original SC-55?

 

 

Edit 2: The DOOM SC-55 recordings by @MusicallyInspired have the more realistic human choir rather than the synth-like/ghostly choir heard in Hexen's CD tracks.

 

Also, I found a recording of DOOM's music being played on a Roland RAP-10 card, and the choir in E1M8 (18:16 in the video) is the ghostly one from the Hexen CD music:

 

 

So it seems that the Hexen CD tracks were not recorded on an SC-55 at all, but rather a Roland SC-7 module or RAP-10 sound card.

 

Interestingly, both Virtual Sound Canvas and Microsoft GS Wavetable Synth seem to use SC-7/RAP-10-based choir samples rather than the more realistic ones heard in SC-55 recordings.

Edited June 4, 2021 by TheUltimateDoomer666

[Danfun64]

Fascinating. Do all of the samples in the Microsoft GS Wavetable Synth that differ from the SC-55 come from the SC-7 or RAP-10?

[TheUltimateDoomer666]

16 hours ago, Danfun64 said:

Fascinating. Do all of the samples in the Microsoft GS Wavetable Synth that differ from the SC-55 come from the SC-7 or RAP-10?

No. MS Synth is a peculiar mishmash of samples from various Sound Canvas models, even more than VSC is.

 

A prominent example is the picked bass which can be heard in E1M1. The SC-7/RAP-10, VSC, and SCVA-55 (SCVA with SC-55 map), along with hardware Sound Canvas models in SC-55 mode, all use the SC-55 picked bass. MS Synth instead uses the SC-88 picked bass.

 

Here is a short comparison using the soloed bass line from E1M1:

 

https://www.dropbox.com/s/gci4lhn8iiegq3b/doom_e1m1_ch3_bass_scva55-vscmp1-msgsws-scva88.mp3?dl=0

 

The examples in order are SCVA-55, VSC VST (from the VSC-MP1 bundle), MS Synth, and SCVA-88. For the sake of consistency, synthesizers other than SCVA were lowered in volume to match SCVA's output level at 127 (according to James-F from VOGONS, SCVA has the same amount of headroom before clipping as the hardware SC-55, so rather than boosting SCVA's volume, other sound sources should be lowered to match SCVA's output level).

Edited June 6, 2021 by TheUltimateDoomer666

[nadiayorc]

This does indeed sound very good, but only when massively boosted in volume, as some others have said. I've been using it in Ableton with Sforzando for various things and I always need to add at least 1 or often multiple +20 dB boost effects, normally around 30dB to get it in line with the rest of the instruments/samples/plugins I use (the max allowed in one instance of the effect is 20) to all the tracks using it. This is the only soundfont I've ever had to do this with, but yeah, it still sounds good when boosted up.

 

It's not exactly a major problem but it makes it a bit more awkward when using the soundfont by having to put extra effects on that I otherwise wouldn't need.

For people just wanting to use it to listen to midi files it might be an even bigger problem if they don't have an easy way to actually boost the volume.

 

Edited June 7, 2021 by nadiayorc

[TheUltimateDoomer666]

The SoundFont is somewhat quieter than the official SC-55mkII recordings from the DOOM Unity re-release, at least when using the SoundFont with CoolSoft VirtualMIDISynth, so I think the presets do need to be edited to be louder.

[RichB93]

Any chance of this being shrunk further? I'm a bit confused to the size, as I thought the SC55 was a 4MB ROM like most of it's contemporaries. Although with that being said, I suppose I've answered my own question in that it probably was 4MB compressed. I'd love to use this on my AWE32, but I don't really want to upgrade the RAM I have, as I'd only really want it to try this.

 

I went through it with some tools and found 13 duplicated samples that are nearly 900k. 

[TheUltimateDoomer666]

2 hours ago, RichB93 said:

it probably was 4MB compressed.

Even when I use SFPack's maximum compression level, the SoundFont only shrinks to 6.88 MB. And that's without the GS variation tones being included; this SoundFont only contains the 128 GM tones (plus the 9 GS drum kits). The SC-55mkII in contrast has an additional 226 tones (though most games don't use them), so including those would likely make the SoundFont even bigger.

 

There are probably a number of reasons for why the size is so large compared to the ROM. One of the reasons is that the SC-55 stores samples in a certain way that reduces size without reducing quality, but is impossible to do with the SF2 format. Gez mentioned this earlier in the thread:

 

On 6/3/2021 at 3:38 AM, Gez said:

I think one of the effects I remember being mentioned was mirroring, where the samples is first played normally and then played in reverse, which when appropriate to use, allowed to only store half of the sample, cutting ROM size requirement for this sample by half. (Something to know about computing in the nineties: every single byte counted!)

 

So unfortunately, I think the only way to get this SoundFont down to a small enough size to be usable on older computers would be to actually reduce the bit depth and sample rate of the samples themselves. This would of course noticeably reduce the sound set's quality, although I think the Virtual Sound Canvas did a similar thing, as Roland managed to cram the Sound Canvas sound sets into pretty tiny files:

• SC-55 | 354 tones + 10 drum kits | 1.72 MB
• SC-88 | 546 tones + 15 drum kits | 2.68 MB
• SC-88Pro | 902 tones + 26 drum kits | 3.78 MB

Edited June 27, 2021 by TheUltimateDoomer666
Additional information.

[THEBaratusII]

Man, I kept reminding @Trevor0402 to make a update post but he tends to forget quite often. I might as well try to bring in the news as best as I can.

 

He got in contact with one of the people who wrote the tools that decrypts/extracts the SC-55 samples and there were plans to on quote "write a tool that prebakes the envelopes into the samples" to potentially make it even more accurate since the SC-55 had multiple envelopes compared to the .SF2 format which had only one envelope. This also could increase the size by a lot and a smaller sized alternative version was to be made for those wanting to use it on their AWE32 cards.

 

Unfortunately, I haven't been seeing progress being made as of late so the best I can say is I might believe the development is either slowed down or on hold.

 

I wanted this to be said because I was honestly worried about the soundfont being abandoned. It really has potential especially since Dwars featured it on his PrBoom Forks video.

Edited July 17, 2021 by THEBaratusII

[Kappa971]

Hi, it's been a few months, is there any news on this soundfont?

[Kappa971]

For those interested, this is where the project was born: https://www.vogons.org/viewtopic.php?f=62&t=76613&start=320

 

They put a lot of effort into it but unfortunately everything stopped and did not lead to the development of an emulator for SC-55 or accurate soundfont :(

[Kappa971]

A new project was born: https://github.com/skjelten/emusc

[OpenRift]

4 hours ago, Kappa971 said:

A new project was born: https://github.com/skjelten/emusc

ooooooooh! Finally!

[brick]

12 hours ago, Kappa971 said:

A new project was born: https://github.com/skjelten/emusc

Awesome news thanks for bringing it to our attention. I'm definitely going to keep an eye on this. I have an SC-55 and I love it, but I'm so worried that it could die on me any day now (30 year old electronics!), I'd love for accurate emulation to be available some day.

 

Mmm I wonder if I can maybe help them test the emulation if they need test samples recorded from a real device to compare, let me see if I can easily extract the ROMs from my devi-

Quote

These ROM files can either be extracted from a physical unit by desoldering the ROM chips and read their content, or you can download the ROM files from the Internet.

... yeah maybe I'll go with the download option.

[NightFright]

Ah come on, who wouldn't ruin their SC-55 hardware for this? Why use the real thing if an emulator is just around the corner...

 

Anyway, personally I decided to spend those 70 bucks for the SCVA and have at least an acceptable SC-55 approximation which doesn't stray too much from the original. It's more than enough for gaming purposes and I'm sure I'll not be like: "OMG the volume of this instrument is so off", "That choir/bass/etc really sucks" etc.

 

If that emulator ever gets finished and it's like the Munt at some point - excellent, I'll retire my SCVA again gladly. At the end of the day, everybody needs to decide for themselves how far they wanna go for their preferred MIDI sound. For some, it's the S-YXG50, the Hyper Canvas, VSC or even the MS wavetable. Whatever makes you happy.

Edited February 20, 2022 by NightFright