How did designers get such usable guitar/pianos/ep/even strings out of tiny sound modules?

[BertKoor]

I had the soundfonts of all the proteus and planet phatt sounds and found them very underwhelming.

Which then begs the question: how was that soundfont put together?
If the soundfont is a 1-to-1 transfer of the original samples (there are some legal issues with that) with all associated parameters (loop points, EG's, etc etc) then you'll be guessing what the f&#% is different.
BUT if some dude just connected up that synth to SampleRobot and rushed that into a soundfont, you know you cannot compare that to the original. It then no longer is an instrument but merely a sample.

Bottom line: it takes a lot of time for very talented people to pay attention to little detail to craft together a very usable instrument with surprisingly few data. It is these limitations and the knowledge you'll have to deal with it.

But anyway, assume you have 4 MB of 8 bit samples on 32 kHz sampling rate. That is a bit over 2 minutes of audio. If you need a good raw waveform: 100Hz would get you plenty of detail, @ 32 kHz that's 320 samples long. Four megabytes divided by 320 equals thirteen thousand different raw waveforms. Be creative, and that surely would be enough to get things done.

So you children are sooo spoiled with yer terabyte drives...

[ras.s]

I just caved in to buying some E-MU soundfonts the other day. I used to have the Proteus 1 device years ago, but gave it away when the screen started malfunctioning (the unit was still operational, though).. The soundfonts are put together by Digital Sound Factory, which is run by the chief sound engineer at E-MU, Timothy Swartz. Apparently he bought the rights to the sample contents and the soundfonts (and Kontakt libraries) are developed directly from them, that is, they're not re-sampled. And I'd expect that a person who worked on the original hardware would have some insight into making them work properly in new formats.

For a bit of background info, look at this Soundbytes Magazine article: http://soundbytesmag.net/dsf-ultimate-e ... iq-bundle/


Must say I'm glad to be reunited with the sounds, even as I can't have the sound of the hardware itself. I'm using the TX16wx plugin to play them back, which isn't too shabby either with its filters, effects and modulation capabilities. Some of the sounds are a bit on the thin side, but with the pack that I got (the one with the sounds from Proteus 1/2/3, Orbit and Planet Phatt) there's still heaps of usable patches. Just add nostalgia. These should be staple sounds for anyone who wants to add a dash of the nineties into their music.


For anyone interested, DSF has constant specials going on and they also do sales quite often (just sign up for their mailing list): https://www.digitalsoundfactory.com/soundfont_products There's also a free pack available, containing instruments from a number of devices.

I hope I don't find myself going down the "got to catch them all" route ...

[JCJR]

The standard soundfont playback engine specs are very basic compared to the parms which the better-quality hardware samplers had. OTOH a soundfont sample set could be imported into a more capable engine, and more control could be applied to make the samples into a "more musical" instrument.

The perception of the "musicality" of a patch might depend on how it is used. If one is just playing a constant-velocity monophonic lead, or slamming out a quantized step-entered three-note chord progression, maybe the differences wouldn't be very obvious. Was most noticeable to me when actually trying to play a patch in real time.

Dunno if the proteus hardware engine inherited the low-level per-sample tweaks which were available in EMU samplers. If the rompler engine contained the same parms, but just hidden below the view of the user, then maybe the samples may have been fine-tuned as good as its gonna get on an EMU sampler before burning to ROM? OTOH if the rompler engine did not have as many low level tweakable parms, then considerably more work would have been required to prepare the ROMs, even if the samples may have been prototyped on a hardware sampler first?

I used to do some of my own sampling on hardware samplers. Before computer audio editors, and then in the early days of computer audio editors. It was like painting a 747 with a Q Tip. First record many versions of each note to digital tape, then either record each note into the hardware sampler, or record it into a computer audio editor, trim, and export to the sampler via scsi or midi. MIDI took forever, transferring audio thru a pipe that slow. Lots of versions of each note were necessary, because even if for instance you might have 10 samples of a D note, each a little different-- It was a job to pick which one of those D notes best-matched the notes you had picked for the C note below it, or the E note above it. Sometimes even with many source notes to pick from, it was hard to get consistent sounds playing up the keyboard. "Gappy" sounding results.

So there was lots of busy work just getting the final sub-samples selected, imported into the sampler, mapped for key range, heads and tails trimmed, etc. Then the real work of looping the samples. Adjust the per-sample amplitude, the per-sample pitch. Some samplers allowed you to individually adjust each sample's filter offset, in attempt to better-match slightly-bright samples against slightly-dull samples. Make global envelope and filter and VCA adjustments, and some samplers allowed you to fine-tweak envelope times on the sub-samples. It took a long time, or at least it took me a long time to do it. One advantage was that since all this tweaking was inside the sampler, you could play the keyboard at any time to judge how it sounds, and so after tweaking enough, you could finally immediately play the final sample to know whether it is "as good as its gonna get".

Am claiming no expertise in sampling. Just that it was a fun thang to do for awhile back in the day, and took some work to get it right. I didn't try to sell samples, but one problem for sound designers who tried to sell the same sound set, compatible with many different brand samplers-- All those low-level tweaking parms were different on every brand, and sometimes different on different models from the same brand. So all that time-consuming fine-tweaking to polish up a sample set, had to be duplicated on each different supported sampler model. Did not seem an enviable way to make a living.

====

And then I fooled around for awhile some years later making Soundfont/DLS samples. DLS standard playback specs are even more stripped-down than Soundfont. Imagine an analog synth so simplified that it only has a handful of knobs that you can twist. It is hard to do much fancy with so few controls.

So the soundfont development HEAVILY depends on sample editing before each sample goes into the soundfont. Maybe there were softwares which integrated everything into one shell. There was a Vienna soundfont editor I recall. But I don't recall it having good enough fine-tweaking of samples. So when I was fooling with it, did most of the editing sample-by-sample in a stereo audio editor. CoolEdit Pro, and maybe some others, can't recall from so long ago.

If you are working on that D note in the audio editor, you can mess with the loops and timbre etc, but its not the same as playing that D note on the keyboard, so you edit in the audio editor, then move the sample to the soundfont and try it out, and then go back to the audio editor again to continue tweaking.

With so few adjustments in soundfont/dls engines, lots of work in the audio editor. For instance short loops-- Bass notes have a long enough wave period that you can semi-reliably find a short loop that will play in-tune with the rest of the sample. But midrange and treble notes have such short wave-periods, if you are limited to integer-spaced loop points, that you can't properly tune a short loop to the exact pitch of the rest of the sample.

So in the stereo editor, with the trusty strobotuner hooked up to the audio output along with the speakers-- First tune the entire sample. Measure the pitch with the strobe (which IMO is about the most precise musical pitch tuning method). With acoustic samples, the pitch will typically move a few cents during note evolution, which is natural and needs to be preserved. But you choose, "do I want all the attacks in tune?" Or "do I want all the sustains in tune?" Or "do I want to split the difference so the entire sample is on average in tune?"

Then micro-stretch each sample so that they all consistently play to your conception of what is in-tune. Then find a short loop that has good timbre, even if the loop ain't gonna be in-tune. After finding a good-sounding short loop, use the strobe to measure its pitch, find out how many cents flat or sharp. Or do some math on the loop length.

For instance A440 frequency, at 44.1 k samplerate, has a period of 100.227272... samples. So if you set a loop length of 100 samples its gonna play sharp, and if you set a loop length of 101 samples its gonna play flat.

So one way to fix the short loop is to turn it into a longer loop. For instance 100.2273 * 13 = 1302.955, fairly close to an integer. So if your "perfect sounding loop happened to be 100 samples, 100 * 13 = 1300. A difference of 3 samples. So you can duplicate/paste that 100 sample loop until it repeats for 13 times. And then stretch-resample that 1300 sample new loop region to a final length of 1303 samples. Now the loop plays very near in-tune, and each individual waveform of the 13-period region has been adjusted 3/13 = 0.2308 samples longer.

So far as I recall, didn't do it exactly thataway. I would just duplicate the loop 16 or 32 times, and then stretch the bigger loop to fit the proper integer length, rounded to the nearest integer sample. Because the loop length error in the original short loop is always less than 1 sample, then after duplicating X32 and stretching, the loop length error of each wave period in the new longer loop is "less than 1, divided by 32" which often gets purt close.

Sorry for rambling. Just explaining that it is a lot of work to get equivalent results with simplified sampler engines.

I do not doubt that those digital sound factory guys do a great job. Merely that it is to me staggering how much work is involved to do a great job on that kind of task.

[dnekm]

Sampling at a higher root note....

Guenter showed me this. Proved to me how to take my Anomaly sounds and turn them into much smaller soundsets.

[planetearth]

I thought I remember reading Dave Rossum saying something about using his custom chipsets to compress the data, and since it was a custom chip it was fast enough to do on the fly. I can kinda buy into that, as I had the soundfonts of all the proteus and planet phatt sounds and found them very underwhelming. My brother has an original proteus and I am amazed at how great it sounds for 4mb(?) of data.

I agree with you on the Soundfont versions (as well as the NI and Dimension versions) of the sounds. If you can use the Proteus VX VST, the presets sound MUCH better.

As Tim (the lead designer on the Proteus synth and VST) explained it to me, the Proteus VX is basically a port of the sounds almost directly from the hardware synth. The other versions (Soundfont, NI Kontakt files, Dimension Pro patches, etc.) are the basic samples with some reverb and/or delay on them. They're really not quite the same.

Steve