It isn't widely known, but I invented virtual instruments. I don't like to brag about it, but I remember the moment of inspiration well. It was winter of 1986 at the Tralfamadore Café in Buffalo, New York, as my bandmates and I sweated, strained, and cursed our way up the back stairway, scraping the walls, bruising our knuckles, and laboring under the weight of a Hammond B-3. Ninety-nine percent perspiration, indeed.
Well, perhaps “invented” is a bit of an exaggeration, especially because my Texas Instruments TI-99/4A wasn't really up to the task of real-time emulative synthesis. Still, my anguished cry of “There has got to be a better way!” resonates through that stairwell to this day.
Seventeen years and countless hernias later, the age of software synthesizers is upon us. In fact, it's in full bloom, and major stars such as Herbie Hancock are touring with virtual substitutes for their signature keyboards. You can now find software emulations of everything from classic analog synthesizers to electromechanical keyboards, drum machines, and more.
I decided it was time to put some of these impostors to the test by comparing them with the instruments that inspired them. I put software and hardware side by side and played them “blind” for a panel of experts with critical ears (see the sidebar “Meet the Panel of Experts”). The results were enlightening.
The scene for the showdown was the Audio Playground Synthesizer Museum (www.keyboardmuseum.com) in Orlando, Florida. Founder and curator Joseph Rivers has assembled what is arguably the world's biggest and baddest collection of keyboards, synthesizers, and drum machines, along with an enormous collection of service manuals, product literature, and other paraphernalia. Think of the most obscure instrument that you've ever encountered; he probably has two of them and can tell you where to find a replacement for the widget that always snaps off in transit.
Clearly, acquiring the reference instruments wasn't going to be a problem. Figuring out how to test them fairly, however, was. At the top of our long list of questions was how to judge the characteristic sound of an instrument that hasn't been manufactured in a decade or more. Should we use a “Mark I” or a “Mark II”? The original version or the “improved” version? For that matter, how much does a 20-year-old instrument sound the way it did originally, and how much did one instrument sound like the next one off the assembly line?
Ultimately, we realized that accounting for all possible variables was simply impossible; we needed to get over ourselves and be practical. We set up the computers and the classics side by side in the Vintage Room of the museum (see Fig. 1), where most of the test instruments were already hanging on the walls. We then adjusted the hardware and software instruments until they were as close to each other as possible and played them for the panel of experts listening in the control room (which, by the way, features 80 synthesizers and more than 1,000 MIDI channels).
The virtual instruments resided in either a Mac G4/733 MHz running a Digidesign Digi 001 or in a Toshiba Celeron 1 GHz notebook with an RME Hammerfall DSP CardBus interface connected to a Multiface I/O box. We split the MIDI output of a controller keyboard to the two computers to ensure consistent MIDI response. Soft and hard instruments all went into the same mixer and were bused from there to the control room. It's worth noting that latency was not an issue with any of the virtual instruments. In fact, we were well into the tests before it even occurred to anyone to think about it.
All of the virtual instruments feature stereo outputs, because none of the hardware instruments did; however, we listened to everything in mono. We also made a point of disabling any built-in effects so we could focus on the raw sounds. Keep in mind that to many users, those features are reasons to prefer the virtual versions, but our priority was to evaluate the accuracy of the emulations.
I was in charge of tweaking and playing, and Rivers sat with the panel, monitoring their observations and recording the examples for posterity. The panelists had only a general idea of the instruments they were to hear, and I identified the instruments by letter or number. We set up a talkback mic, and occasionally, the panelists would ask to hear an example again or to hear a different musical figure played. In rare cases, they made fun of my wrong notes.
IN THIS CORNER …
Our first bout was between a Yamaha DX7 and Native Instruments' FM7 (see Fig. 2). We thought that would be the easiest place to start, because we could match the factory presets that Native Instruments ships with the FM7 against cartridges of DX7 factory patches. In some cases, the match was startling, but in others it was hardly a match at all. Getting to the source of the differences would have been an interesting but time-consuming endeavor, so we declared the glass to be half full (more than half, actually) and singled out the best matches for comparison.
I played a range of sounds, including an electric piano, a synth lead, a harpsichord, and a flute. First I played the FM7 (test A) and then the DX7 (test B). Although there was some discussion over the proper use of the term warm (musicians!), the panelists all felt that the DX7 had more midrange presence, especially in the electric-piano sound. In contrast, the FM7 had more harmonic activity in the sound and was somewhat brighter.
Electronic Valve Instrument guru Sam Zambito summed it up this way: “I concluded that A was the hardware instrument; but I was hoping it was the virtual instrument, because I liked the sonic detail.” That theme was repeated throughout the day as panelists inevitably considered which sounds they would like to have in their personal arsenals. In general, they were pleased that the virtual instruments had sufficient character.
The FM7 has a digital slider that lets the user reduce the bit depth of its output to match that of the original instrument's output. According to Native Instruments, “Unlike all the other features of the DX sound engine, which the FM7 emulates exactly, the quantization noise characteristics are only approximated.” At a setting of zero, there is no bit reduction, and at a setting in the 10 to 15 range, the noise is said to be close to that of the DX7. Armed with that bit of information, I made some adjustments and played the electric piano again. At a setting of 14, the panelists felt the sounds were much more closely matched, suggesting that the digital slider does a good job of making the emulation more convincing.
I also noticed that the FM7 was responsive to key Velocity. That's one of the things I liked most when I first checked out the instrument. The responsiveness not only makes it easy to play colorfully, but it also makes it easier to distinguish the real from the virtual. To compensate, I set up the MIDI Out of the DX7 to trigger the FM7 and selected DX7-style Velocity response (0 to 100 instead of 0 to 127) in the FM7's preferences. That made the electric piano sound almost indistinguishable between the original and software instruments.
At its best, the FM7 was capable of near-perfect emulation of the DX7. Where our panel heard differences, they generally preferred the FM7's brighter, richer sound. Typical was Rivers's comment that the FM7 often “sounded brighter and better to me for today's recordings.”
AND WEIGHING IN AT …
Despite the begging and pleading of roadies everywhere, the behemoth Hammond B-3 has found its way into practically every genre of popular music during the past four decades. It has also been sampled and simulated in a variety of ways in an attempt to capture the famous sound without the lugging. We pitted a real B-3 against two contenders: Native Instruments' B4 (see Fig. 3) and Emagic's EVB3.
The very things that organists love about the B-3 — the nearly infinite combinations of stops, percussion, and vibrato, as well as the always cool Leslie rotary speaker cabinet — make accurate comparisons almost impossible. Rather than spending a solid week testing every variable, we opted to limit the subjective testing to single manual sounds without the more exotic extras.
The first test was with EQ, percussion, vibrato, and other parameters disabled and all the stops fully out. With that setting and after carefully matching volumes, I played some scales, arpeggios, and simple melodies on each of the three instruments. The first thing that jumped out was that with identical settings, the EVB3 was noticeably brighter than the other two. I compensated by trimming some of the higher stops back a bit, and although that helped somewhat, the EVB3 continued to stand out as a generally brighter sound. Qualitatively, the panelists didn't feel that was necessarily a bad thing but merely a difference that made it identifiable.
The B4 came close to the B-3 at the same settings, but there was something about the warmth of the real B-3 that made it easy to pick out. As I adjusted the stops to reshape the timbre, matching the sounds as closely as I could, nobody was fooled. The B4 got compliments for its clear sound, with the B-3 being regarded as the darkest of the three.
Rivers asked me to play a series of single notes on each organ to judge the attacks. “Because the tonewheels of a real B-3 are constantly turning, the attack of a note can come at any point in the waveform, lending variety to the articulation of the instrument,” he says. That was yet another subtle characteristic that separated the real instrument from the software versions.
Throwing pure subjectivity aside, we explored some of the presets on the B4 and EVB3 to see how they sounded at their best. Even on wildly different settings, it was easy to identify each organ. They all had their own characteristic sounds that shone through regardless of their specific parameters. The EVB3 got high marks for its grit and edge, whereas the B4 continued to sound a little more authentic.
I got sneaky on the next comparison by throwing in two Rhodes electric pianos to compare with Emagic's EVP88 and Applied Acoustics' Lounge Lizard EP-1 (see Fig. 4). That caused our esteemed panelists to doubt themselves as they tried to identify which instruments were originals. Interestingly, they weren't so confused that they failed to identify at least one of the real pianos as being nonvirtual. Rivers identified the Rhodes 73 as the real one and then, assuming that only one was an original, declared the 88-key Stage Piano Mark I to be “the best virtual instrument I've ever heard.”
The other interesting thing about the electric-piano comparison is that even though they could confidently pick out the original ones, nobody cared. “I'd be happy playing any of those pianos,” says Per Danielsson. The other panelists were similarly pleased with the virtual pianos.
Rivers asked for a crescendoing series of single notes on each piano, and we noted that the real ones exhibited more bark as the notes were struck harder. Zambito observed that the real pianos also had more tine distortion, which resulted in a more varied attack compared with the clean and consistent attacks of the virtuals.
Meanwhile, behind the curtain, I had some observations about playing the four pianos. The first thing I noticed was how mushy the “ancient” keyboards of the two Rhodes instruments were. Even though I was playing a light semiweighted MIDI controller, the virtual pianos were more comfortable to play. Had the panel members heard the resounding clunk of the Stage Piano's damper pedal, they would have had even less doubt about which one was hardware. Score one for virtual instruments.
However, when I played the Lounge Lizard with the damper pedal down, we noticed that the instrument's note-allocation implementation could lead you into trouble. I played an arpeggio with the pedal down, and when I reached the maximum polyphony, the first notes gave way to the new notes, as you would expect. Then I played an arpeggio and repeated the top note several times; the same thing happened. Apparently, each time you strike a repeated note, it eats up a note of polyphony, eventually causing the lower notes of the arpeggio to disappear one by one, leaving only the repeated note.
We compared that with the EVP88's behavior and found that the EVP88 handles that situation much more gracefully, recognizing the repeated note and preserving the other notes. On the Lounge Lizard, you can avoid the problem by holding down the lower keys with your left hand while repeating the upper note. Out of curiosity, we tried the same experiment with the FM7 and found that it sacrificed the oldest notes regardless of whether you held the keys down by hand or just used the damper pedal. We all agreed that was the least attractive behavior. Score one for hardware and the EVP88.
NOT FOR PROPHET
One of the most challenging comparisons (for me, anyway) was the Sequential Circuits Prophet-5 against the Native Instruments Pro-52. For those two, I had to choose some Pro-52 patches and then attempt to match them by hand on the Prophet-5. I started by visually matching the positions of the numerous knobs and then adjusted them by ear until I had a good match. Fortunately, the Prophet-5 does have user memory, so I could save several patches and switch between them without delay for the demo.
The Prophet-5's keyboard is not Velocity sensitive, so I had to be careful that I didn't give away the Pro-52. My MIDI controller had a fixed Velocity mode, but it made the Pro-52 sound dull and dark. Apparently, the keyboard sent out a fixed Velocity value of 64 rather than 127, triggering the Pro-52 in the middle of its response curve. It was much easier to just hammer the keyboard to ensure I sent all 127s. That technique matched the Prophet-5 much more closely.
The Prophet-5's keyboard also had some bad keys, so I had to play everything in keys such as F-sharp and D-flat, taking advantage of the less-used black keys. Score another one for virtual instruments or at least for instruments less than a zillion years old!
The consensus on this matchup was the opposite of the DX7/FM7 pairing. The panel found the Prophet-5 sounds to be brighter, bigger, and more open than the Pro-52. As composer and bassist Andy Hagerman said, “It wasn't that the virtual synth sounded bad. But I was hoping for it to be the more interesting one, because I'm not terribly interested in buying and maintaining an antique keyboard. Unfortunately, the classic has a deeper, more interesting sound.” Danielsson noted, however, that the Pro-52's attack was smoother than the Prophet-5's and that its sound was nice and solid.
The panel wondered what would happen if we started turning knobs on the two, suspecting that the software instrument might reveal its digital nature in such a test by producing a grainy or “zippered” response. On the contrary, the panelists were quite pleased at the smoothness of the virtual filter sweeps. The only giveaway was that the sweeps were less even because I was controlling them with a mouse instead of a knob. All of the Pro-52's parameters respond to MIDI controllers, so getting around that limitation would be simple with an appropriate keyboard or control surface.
INTERNET, MEET CLAVINET
Our final contestant was Emagic's EVD6, an emulation of a Hohner D6 Clavinet. When we were setting up, we noticed that the real Clav had an extraordinary click on the release of each key. The click is a normal part of the Clavinet sound, but this was huge. Rivers suggested a couple of ways to hide the click so it wouldn't be so obvious that it was the real deal. As soon as he left the room, I dug into the interface of the EVD6 to see if I could make the virtual Clav as funky as the physical one. Sure enough, by maxing out the Click parameter (see Fig. 5), I was able to get a satisfying release.
By the time we got around to the Clav test, only Zambito was left from our original panel of three, so he and Rivers had to endure my rendition of “Do You Believe in Love?” by themselves. Both were quite certain that the first instrument I played was the original D6, and in fact, they were correct. When I played the EVD6 for them, however, they shared a moment of uncertainty because of the realism of the click. In the end, it was primarily the evenness of the EVD6's click that made it sound too well behaved. (In my enthusiasm for the magnitude of the click, I overlooked the “random” slider right below it.) Still, the realism of the sound elicited the only “Wow!” of the afternoon.
We noticed that the D6 rang a lot longer on sustained chords, and in about ten seconds, I had adjusted the sustain time of the EVD6 to match. As we dug deeper into its interface, we discovered that it not only duplicates the damper slider of the Clavinet and its multiple pickups but also goes way beyond the sonic possibilities of the original.
None of us was terribly surprised to find that we could consistently identify the original instruments. Given the differences in the electronics between the instruments and the computer output stages alone, we had expected to see some telltale signs. Nevertheless, in most cases, we were able to adjust the virtual instruments into a close — occasionally even remarkable — likeness of their namesakes.
Each of the virtual instruments that we tested does a good job of emulating its hardware equivalent. As Hagerman says, “I doubt even geezers like us would be able to pick out these differences in anything other than this A/B comparison.” More importantly, the panel of experts felt that the software synths were viable instruments in their own right, often going beyond the emulative aspects to extend the sonic legacies of their forebears.
Although for the purposes of our tests, we were primarily concerned with the accuracy of the emulations, the virtual instruments also offer many conveniences not found in the originals, such as unlimited patches, computer-based editing, and MIDI control of things that only a technician could change in hardware. There is indeed a better way, and musicians and roadies alike have a genuine reason to celebrate.
Brian Smithersis course director of advanced audio workstations at Full Sail Real World Education in Winter Park, Florida. Thanks to Andy Hagerman, Sam Zambito, Per Danielsson, and Joseph Rivers for their invaluable assistance.
MEET THE PANEL OF EXPERTS
Keyboardist Per Danielsson teaches jazz studies at the University of South Florida in Tampa and has two CDs that are both in the final stages of production: one featuring his trio, and the other featuring a collaboration with drummer Danny Gottlieb.
Andrew Hagerman (www.singularityarts.com) is a bass player, a composer, an ILDA Award — winning producer of LASER shows, and an associate course director of advanced audio workstations at Full Sail Real World Education.
Engineer, producer, and synthesizer authority Joseph Rivers operates the Audio Playground studio and its world-famous Synthesizer Museum and has worked as a consultant and sound designer for major synthesizer manufacturers.
Trumpeter Sam Zambito (www.samzam.com) is one of the first and most prominent proponents of Nyle Steiner's Electronic Valve Instrument and has done sound design and programming for the likes of Michael Brecker and Bob Mintzer.