Byline: Brian Smithers
They say the more things change, the more they stay the same. When it comes to synthesis, that statement certainly seems to be true. Names such as Moog and ARP come up in conversation as often as they did 30 years ago. This time around, however, the discussion is tempered by decades of improvements and conveniences in synthesizer design, such as patch memory, MIDI controllers, and computer synthesis.
The era of software synthesizers is in full bloom. Although many original synthesizers are being built entirely in software, numerous others are intended to emulate classic instruments from days gone by. EM decided it was time to check the state of the art in synthesizer emulation, so I tucked a computer under my arm and headed off to the synthesizer retirement home (yeah, I live in Florida) to do some comparisons.
The retirement home in question is the Audio Playground Synthesizer Museum in Winter Park, Florida (www.keyboardmuseum.com). Founder and curator Joseph Rivers has assembled an awe-inspiring collection of synthesizers and drum machines, from the classics to the curiosities. In actuality, retirement home is an inadequate description because the museum is housed in a full-fledged modern recording studio and features many of the latest synthesizers right next to the oldies.
Matchmaker, Matchmaker
The matchups for our comparison consisted of an ARP 2600 with Arturia's 2600V and Way Out Ware's TimewARP 2600; a Minimoog with Arturia's Minimoog V (and a Minimoog Voyager just for fun); a Roland TB-303 with Muon's Tau Bassline Mk2; a Korg MS-20, a Polysix, and a Wavestation with their counterparts in the Korg Legacy Collection; and a Yamaha CS-80 with Arturia's CS-80V.
Our goal was to find how closely the virtual instruments sounded like their namesakes, therefore blind comparisons were in order. It was immediately apparent that numerous challenges had to be addressed. Some older instruments clearly gave themselves away by the level of hum and other noise they produced even before I played a note. To minimize the prejudicial effect of that unavoidable reality, we kept the real instruments live while playing the virtual instruments, so the virtual ones were heard with the same background noise as the real ones.
One of the most challenging parts of the comparison process was trying to ensure that any bonus features of the virtual instruments didn't leave telltale signs. I played everything in mono, turned off all effects, watched for unison modes where none originally existed, and matched polyphony carefully. Some virtual instruments made that easier than others. For example, Arturia's documentation usually identifies virtual-only features clearly for those obsessed with veracity.
Another noteworthy aspect of the testing was that the wilder the sounds got, the harder it was to make comparisons. For one thing, many of the (usually excellent) presets that ship with the virtual instruments under our microscope make a point of using built-in effects such as chorus and delay and using extra oscillators or modulation matrices. Turning those features off often robbed the patch of its essence, so it was often not useful to start a comparison from a preset. Because of that, in some cases the comparisons start with a single oscillator and build from there.
Similarly, the range of certain controls was often quite different between the real and virtual instruments. That could be attributable to age-related drift in the real instruments or to shortcomings in the design of the virtual instruments. In some cases it was possible to match sounds closely by using different settings, such as a filter cutoff set significantly lower and with much less resonance on one instrument than on its counterpart.
The first round of comparisons, which included the 2600s, Minimoogs, TB-303s, and MS-20s, took place in Audio Playground's Studio B, which fittingly features more than 1,000 active MIDI channels connecting 80 or more keyboards and synth modules. Of course, given the vintage synths under examination, we used none of those channels.
I set up in the producer-performer area at the back of the room, and the esteemed panel of experts (see Fig. 1) sat at the console. An improvised screen shielded my activities from their eyes while allowing easy communication. Occasionally, panelists would ask to hear an example repeated or suggest a variation. The entire session was recorded in Pro Tools, and Web Clips are available online for most of the comparisons.
To keep the playing field as level as possible, all but one of the virtual instruments was played from the Open Labs OpenSynth neKo 64 keyboard workstation at 24-bit, 96 kHz resolution. The Way Out Ware TimewARP 2600 was available only as an RTAS plug-in at the time (AU and VST versions are in the works), so it was played through a Digidesign Mbox at 24-bit, 48 kHz resolution. Real and virtual instruments were patched to the console in mono and panned dead center. Levels were hand-matched carefully and often had to be changed from one patch to the next.
The first round was besieged by a number of time-consuming difficulties with the vintage instruments (score one for the virtual instruments), so a second round had to be completed remotely. I returned to the museum to record the comparisons for the Wavestation, CS-80, and Polysix, and I posted uncompressed mono files online for the panelists to analyze.
As the operator, I have to confess an unbridled preference for the virtual instruments. In fact, after spending many hours wrestling with drifting oscillators, sticky sliders, noisy outputs, faded silkscreens, persnickety connectors, and other assorted electronic maladies, I felt the need to spend a good solid week playing my bamboo flutes barefoot high in a tree just for balance. Certainly enough time and money would make playing the vintage instruments less like playing Russian Roulette, but for me the cost-benefit analysis weighs heavily in favor of spending that money on a fast CPU and a low-latency audio interface so I can spend my time making music with the soft synths. Our priority here is the accuracy of the emulation, however, so let's see what the guys with the golden ears thought.
Be wary of drawing inferences of superiority based on descriptive terms such as warmer or brighter. We each had preferences here and there, but they were often based on analyses more complex than isolated timbral distinctions.
ARP 2600
The first synth under the microscope was the first synth to cause problems. Although the museum's ARP 2600 functioned pretty well, its companion keyboard didn't. To be fair, it wasn't clear whether the keyboard or the synth was at fault, but there was no way to get the two cooperating well enough to play musical phrases. The ever-resourceful Rivers brought out a MIDI keyboard with a MIDI-to-CV converter, and that was better, but still not up to the task. We had to resign ourselves to comparing the oscillators and filters more like scientists than musicians.
After carefully matching output levels between the 2600 V (running on the neKo 64), the TimewARP 2600 (running on the Mbox), and the ARP itself, and doing everything possible to eliminate LFOs and filters as variables, I played raw waveforms for the panelists. Starting from silence, I ramped up the level to full volume, held it there for a few seconds, and pulled it back down. I immediately repeated the process with the other two instruments. Some differences revealed themselves immediately. I had expected that moving the onscreen control with my mouse would make the TimewARP (see Fig. 2) an obvious virtual, but in fact it was the ARP that gave itself away with a sticky slider. On subsequent examples, I was able to coax somewhat smoother behavior from the ARP.
In comparing sawtooth and pulse waves, the panelists had no trouble identifying the real 2600, although not necessarily for the reasons that one might expect. Bassist and composer Andrew Hagerman declared all three instruments "astonishingly alike except for some subtle color differences." The 2600 V's sawtooth was unanimously declared brighter than the others, with the TimewARP's sounding rounder, fuller, and more interesting. The 2600 V's pulse wave sounded fuller than its counterparts, which were described as nasal by comparison.
Interestingly, when I swept the 24 dB-per-octave lowpass filter on each instrument, the most satisfying result for all of the panelists was from the TimewARP, with the ARP exhibiting a much coarser behavior than the others. Based on that, three of four panelists felt the TimewARP sounded more real than the real ARP. The ARP's cutoff control was jittery enough to be described as bad digital stair-stepping. The 2600 V also sounded stepped to three panelists, albeit less so than the ARP.
The ARP's sticky pot made me wonder about the differences between the physical controls involved. The Arturia was being controlled from the neKo 64's touch screen, whereas the TimewARP was being controlled from my notebook's Accupoint pointing device. (Accupoint is Toshiba's term for the eraser-head-style pointing stick, which I have always preferred to the more common glide pad. Now I have one more reason to prefer it!) It is hard to gauge the extent to which the difference was attributable to the physical control, but it's important to note that the expected superiority of a physical control clearly diminishes when that control is 30 years old.
For the final 2600 example, I cranked up the resonance and swept the filters again. The resonance characteristics varied more widely than any other parameters we tested, with the real ARP self-oscillating earlier in the control's range and more wildly than either virtual. That generated the only "yikes!" of the evening. The TimewARP again scored points for smoothness, with the Arturia displaying interesting artifacts, which were regarded positively by one panelist and negatively by another.
Korg MS-20
You would expect the instruments of Korg's Legacy Collection to be dead ringers for the originals, but the MS-20 was tough to match. Still, it managed to cause a bit of confusion. Two technical challenges, one analog and one digital, caused major delays in preparing the examples. The first real MS-20 we tried had such bad drift in the oscillators and in multiple components that by the time I got the virtual instrument matched, the sound of the real one had changed dramatically. Fortunately, the second unit was better behaved.
I had hoped to use the Legacy Collection's MS-20 Controller (see Fig. 3) - a USB device slightly smaller than the original MS-20 that reproduces its controls right down to the patch cables - to simplify the setup process. I quickly discovered, however, that settings on the controller don't always line up very well with the settings in the software. For example, raising a knob from a setting of 0 to 1.5 often failed to move the onscreen control at all, and higher settings were off by one or more values often enough to make the controller more hassle than it was worth in this context. (In performance, of course, you're usually not trying to match the settings of another MS-20 on the fly, so this is not an indictment of the controller when used for its intended purpose.)
One of the nice touches about the Korg Legacy Collection is the inclusion of the original manuals on the installation disc, including the Setting Examples, consisting of patch documents to be used as examples and blank patch sheets for user settings. I started with a couple of the example patches, first dialing them up on the real MS-20 and then matching the settings on the virtual instrument. Matching the knobs as carefully as possible did not result in a convincing sonic match. With a bit of imagination and effort, however, I was able to get the two to sound much closer.
The first patch was labeled Trumpet in the Setting Examples, and the drifting oscillators seemed to give the real instrument away. Despite its erratic pitch, it was described by synthesist and trumpet player Sam Zambito as "vibrant, rich, and detailed," with a "stronger character" than the virtual version. The rest of the panelists agreed, calling the real instrument "more substantial" and "beefier" and the virtual instrument "thin" and "too clean."
On the second patch, the comments ran along similar lines, so I decided to experiment with the virtual MS-20's Analog knob. Like some other designers, Korg has decided to allow users to determine how much old-school random behavior they want their soft synths to exhibit. I cranked the knob way up and reversed the order of the examples, and all of the panelists were convinced that the real MS-20 had simply gone further out of tune. Thus I scored the first and only successful deception of the evening.
The real MS-20 generally sounded bigger and richer than its virtual cousin, but as Hagerman put it, "Do I really want to battle drifting oscillators to get a slightly more present sound?" It was difficult to get rid of a persistent ensemble sound in the virtual MS-20, a characteristic that was not disliked but was taken as a digital giveaway. The most convincing part of the emulation was the analog=Misbehavior knob, a mixed blessing outside of our context.
Moog Minimoog
The Minimoog (see Fig. 4) was one of the highlights of the shootout, both for the relative good behavior of the real Moog and for the quality of Arturia's emulation. There was actually a split decision from the panel on which was which, validating the accuracy of the Minimoog V's sound.
I started with a one-oscillator patch, playing a few examples and varying the patch slightly as I went. Rivers used the term "rich" to describe the real Minimoog, while Zambito used the same word to describe Arturia's virtual version, demonstrating once again that musicians don't adhere to the same descriptive standards as scientists. Their description of the real Minimoog's high end was more revealing - one panelist said it sounded "more open," and another said it had a "thinner" sound.
Three panelists independently used the term "reedy" to describe the Minimoog's sound as I tweaked the filter a bit. I tried hard to emulate that quality in the virtual version, but I experienced only a limited amount of success. There was always a bit of grit in the Minimoog's filter that the virtual one couldn't quite replicate. On the other hand, our experts agreed that at times, Arturia's virtual version sounded "dense" compared with the real Minimoog. At least one patch was described as better able to cut through a mix than the real instrument.
I couldn't complete my comparison of the Minimoog without trying my hand at a bass patch, and I took the liberty of switching the order just to keep everybody honest. Everybody knew immediately that I had switched the order; not everybody, however, picked which instrument was which correctly. Although the instruments were distinguishable, the entire panel thought the Minimoog V captured the sound of the original well enough that, as Hagerman put it, "given the extra functionality and consistency of the software, I'd probably opt for the soft synth on practical grounds." Rivers added, "great software - I think they did the Moog justice."
The Museum counts within its collection a signature-edition Minimoog Voyager, and we couldn't resist matching it against the others, so I dialed in the bass patch as closely as I could. This test was not blind, and we had already discussed the other two in detail, but it was nonetheless interesting to hear. Composer-keyboardist Lee Riley described the Voyager as "analog by nature, but with a clean digital approach - like Sean Connery in an Armani suit." He went on to opine that the Minimoog V came closer to the Voyager than to the original Minimoog. Zambito felt the Voyager "seemed to deliver the best attributes of the [classic] Minimoog and virtual synth."
Roland TB-303
Muon's Tau Bassline Mk2 (see Fig. 5) models the synthesis, but not the sequencer, of the original TB-303, so I had to play both by hand. Given the tiny buttons that pass for keys on the 303, it took some doing to match the performances. Once I did, however, the differences between the real and virtual almost completely disappeared.
"Wow" appears in Hagerman's and Rivers' notes, with Zambito concurring, "great!" Riley and Zambito found the Tau "thicker" and "richer" than the original, but all agreed that the two could be made to sound almost identical. At one point I simply played alternating notes on the two, and it was difficult to tell that they weren't all coming from the same instrument.
Getting to the same place on both instruments took several minutes because the range of the cutoff and resonance controls on the two overlapped but did not completely coincide. As Hagerman put it, "The soft synth has a bit more range of expression and a bit more edge, but [at best] they are so close!" Score one for the fine art of doing one thing and doing it well.
Korg Polysix
The Polysix (see Fig. 6) was well-behaved except for being slightly flat, even when tuned as high as it would go. The variability of its controls, however, made for some challenges in trying to match sounds. If you've ever tried to hang a picture and found that every time you tried to level it you over- or undercompensated, then you've got an idea of the difficulty that I had. (The original Polysix has 32 user memories and tape backup, and thank goodness it does!)
As with the MS-20, I started with some of the example patches from the original Polysix settings manual. A brass patch with a delayed pitch modulation sounded similar on the two, but there was a pop in the attack of the original instrument that I couldn't reproduce on the Legacy Collection virtual instrument. Conversely, there was a nuance in the virtual instrument's attack that didn't seem achievable in the vintage version. Ideally, I would have liked to get both attributes at the same time. Composer Riley noted the warm ensemble character of the virtual version, and indeed the real Polysix sounded timbrally flat by comparison, contrary to our biased expectations.
Next I took a preset from the soft synth and tried to match it on the hardware instrument, with mixed results. In this example, the software revealed itself by its consistency from phrase to phrase and from note to note. Hagerman felt that it had "a great core to the sound, and a constancy that makes me say that it's the plug-in." The hardware had a long-cycle modulation that took the heart out of the sound periodically despite my best efforts to remove all modulation sources from the equation. It sounded similar to the software at the brightest end of the cycle, but its variability was telling and frustrating.
In setting up a pizzicato bass patch, it turned out that getting a timbral match was not nearly as difficult as getting an articulation match. It felt as though the real instrument needed a fine-tune knob to go with each of its envelope controls because the smallest turn of the knob seemed to move from too short to too long. It seems almost sacrilegious to say that it was much easier to use the mouse than the real knob, but in this case that was true. Bass player Hagerman felt the soft synth lacked the weight of the hardware, sounding "a bit sterile." He concluded, "I'd choose the second device [the real Polysix] in a heartbeat, and layer it with another sound that can give me a little more attack; then I'd have the best of both worlds!"
To test the behavior of the filter, I experimented with a tinkly space-harpsichord until I got a near match. The Polysix was noticeably out of tune from octave to octave, so I was restricted to playing narrow-range examples. Even then, the real instrument was revealed by its drifting oscillators. Within that range, the basic timbre was reasonably easy to match, but the characteristic of the filter's resonance was quite different between the two instruments. The vintage instrument's resonance was more often dissonant, whereas getting a pleasing twang out of the virtual instrument's resonance was easier. Riley and Hagerman came down on opposite sides as to preference. Riley felt that the virtual "had trouble with the subtle nuances inherent" in the real instrument's sound, whereas Hagerman found the pitch problems distracting, preferring the "bit of extra shimmer" in the virtual instrument's timbre.
For the final Polysix example, I once again pushed the Analog knob on the virtual instrument to see what happened. In this case it took an organesque patch and grunged it up quite successfully. Ultimately, it made the soft synth sound older and more messed up than the old Polysix.
Yamaha CS-80
Yamaha's CS-80 features two layers of push-button presets, but those presets predate ROM-based memory, instead recalling settings stored on small sets of sliders that duplicate the main sound controls on the face of the instrument. That makes the presets stored in a vintage instrument unreliable as indicators of what the instrument's designers really had in mind; they still, however, provided a convenient point of reference for our comparison.
The CS-80 (see Fig. 7) proved to be one of the hardest instruments to match with its virtual counterpart. I spent a good deal of time trying simply to re-create a patch on one that sounded good on the other, and I came away frustrated every time. That is not to say that the virtual instrument doesn't sound like the original; it does, but not in the carbon-copy sense. Compared to this particular vintage CS-80, the CS-80V is more of a fraternal twin than an identical twin - they are clearly born of the same lineage, but not of the same DNA. After hearing several examples, Hagerman concluded, "This might not be the best example of a plug-in copying a vintage model, but it certainly captures the spirit."
That qualification is no small factor in a comparison such as the one we are doing for this article. It's impossible to tell how much a particular specimen sounds like it was intended to sound 30 years ago. If you fell in love with the sound of your friend's classic CS-80 and then found one of your own on eBay, could you realistically expect them to match each other perfectly?
Each of the examples started with the presets, most of them a 50/50 mix of a Part I preset and a Part II preset. From that point, I tried to eliminate variables and match filter settings for the best timbral affinity. In general, the real CS-80 enjoyed a deeper bass response and meatier filter modulation than the Arturia version, although the range of timbres and amount of modulation control offered by the virtual instrument were impressive.
The Strings 1 and 2 presets for the virtual instrument didn't sound very close to those of the real instrument, but layering Strings 3 and 4 produced a better match after some filter tweaking. The low end of the real CS-80 had more of a core to the sound, but both instruments achieved the thick, sinewy texture that one would expect from a brass patch, despite the name. The panel heard a bit more "Aftertouch" in the virtual instrument, along with "a lot more of that traditional analog fuzz."
Next up was the Flute patch, which exists only on Part I, so I mixed Part II out completely. The Arturia version of that patch was respectably flutelike, whereas the Yamaha version was more nasal and variable over the course of a phrase - an interesting but pretty strident effect, and less like a real flute.
There are two pairs of Funky presets on the CS-80, and both made for interesting comparisons. Funky 3 and 4 use a big, slow filter sweep that made for a dynamic fanfare. The real instrument again sounded a bit fuller on the bottom end, but the virtual achieved a more intense edge at the peak of the filter sweep, teetering on the brink of self-oscillation.
Funky 1 and 2 sounded like a bass line to me, so that's what I played. The Arturia had a nice touch wah sort of attack that wasn't present in the Yamaha, whereas the Yamaha almost sounded as though it had a subharmonic generator fattening the bottom end. Once again we turned to Hagerman, our resident bassist, who said, "Personally, I prefer the [Arturia], but based upon any given project's needs, I might choose either."
The final test of the CS-80 was an equal layering of Organ 1 with Organ 2, and the results were in line with the previous examples. The Arturia CS-80V was easily distinguished from the Yamaha CS-80 by its clarity and evenness. Nevertheless, all of the things that are interesting about the original - richness and variety of sound coupled with impressive modulation possibilities - are present in the emulation. Its additional modulation matrix, effects, and layering capabilities make it even more distinct from the original.
Korg Wavestation
The Wavestation (see Fig. 8) made for a particularly interesting comparison, pitting two digital devices against each other. Unsurprisingly, that made for some of the most perfect matches of the entire session. Still, some patches sounded different from real to virtual instrument.
The real instrument was the rackmount version, the Wavestation A/D, which slightly predates the Wavestation SR modeled in the Legacy Collection. The A/D's joystick controller, used to control the Wavestation's Advanced Vector Synthesis, is quite small, giving the neKo 64's touch screen or even a regular mouse a bit of an advantage for subtle timbral control. It was also convenient on the virtual instrument to be able to Ctrl + click to center the control or double-click on a corner to jump to that spot.
I paired up the first wave sequence that caught my ear, called The Wave Song. It was nearly impossible to hear any difference between the real and virtual versions. Being a decade or so newer than most of the other synths that we heard and not being dependent on lots of knobs and sliders to create its sounds, the Wavestation produced the best emulation right from the start. Riley commented, "Wow, this is a toughie. These two samples are pretty much identical." There was no issue of excessively noisy output or oscillator drift, and the digital nature of the PCM waveforms and the patch information gave us a true apples-to-apples comparison.
