Audio Modeling SWAM Instruments – the Synth and Software Review

In which the reviewer gets totally blown away (figuratively, and quasi-literally using a wind controller)

Having to pull myself away from playing the Audio Modeling SWAM instruments to report on them is really tough!

These instruments have been around for a while, but recently they came out with string sections to join the solo strings, woodwind, and brass instruments they’ve been developing all along. It’s time to check in and see how far their technology has come, in particular how it compares to sampling for acoustic instrument emulation.

For the TL;DR set, we quote George Takei and the late Dick Enberg for a two-word summation: oh my. This is the most excited I personally have been about a musical product in years (in this category, anyway). These instruments are just extremely satisfying to play.

Before getting into the details, we suggest you go to the Audio Modeling site and try a working demo of the SWAM instruments. Also note that there are several short audio demos scattered throughout this review (because of a background as a recorder player, you’ll have to excuse a lot of my off-the-cuff noodling for tending to take on a a baroque character).

Sampling technology has strengths and advantages, starting with the inherent life good libraries have because… well, because they’re recordings of live musicians. With the SWAM instruments it’s up to you to create the life with your performance. The good news is that it’s not especially difficult to do that, and they respond with expressive nuances to your playing.

Check out the following example of a SWAM double bass noodle that I just tossed off in real time. It may be a little agile for a real bass, but please focus on how long it would take to edit a comparable performance using samples – if it were even possible.

[UPDATE: After getting some flack online about the bass example here’s an interim link to the same MIDI an octave up, playing the SWAM viola. Unlike a bass player, a violist could play this no problem. – NB]

---

---

I’m using an Akai EWI5000 wind controller, but you can use any instrument that sends MIDI continuous controllers – pedals, sliders, aftertouch, a ribbon controller… That instrument can be a regular MIDI keyboard, or instruments like a ROLI Seaboard, Linnstrument, Haaken Contiuum… you name it. There’s a screenshot later to show how every SWAM instrument comes with presets for most of the controllers out there.

You could use a stage piano-type keyboard – that is, one that doesn’t even have pitch and mod wheels, let alone aftertouch – but that’s like saying you could swallow cherry pits and throw away the fruit. Why would you want to?

Behavior modeling

The fundamental feature of all the SWAM instruments is their “behavior modeling” – the way they respond to your playing like their acoustic counterparts, in astonishing detail. If you “blow” hard and quickly into a SWAM flute, for just one example, you’ll get a chiff sound – even if you’re already sustaining a softer note. Furthermore, the chiff isn’t on or off – the transition between a normal tone to an overblown one is continuous.

---

Audio Modeling gets the feel right too. A piccolo is slightly more agile than a bass tuba, and that difference is incorporated in the instrument models. And by the way, that SWAM bass tuba will shake your house. It’s absolutely wonderful.

---

Similarly, if you have the nerve to go for a low B-Bb trill on a trombone… okay, SWAM won’t deliver the swift kick in the ass a real trombone player would give you for making him or her do speed bag routines with the slide, but the instrument will glitch realistically. The interface always has an informative animation of the instrument you’re playing, in this case showing the trombone slide positions jumping around into position as you play. 

Or you can set the Play Mode to valves – like a valve trombone – and then you’ll see the fingerings animate in real time. With the string instruments you see the bow moving back and forth on the string it’s playing, the correct distance from the bridge, along with the finger position. The winds don’t show fingerings, but they have what seems to be the air moving through the tube at different pipe lengths.

Comparison to sampling

Before GigaSampler came out just before the turn of the century with disk-streaming technology, sample libraries were limited to the size of your sampler’s available RAM. Removing that restriction allowed instruments to be sampled in as much detail as necessary, which made emulations far more realistic and expressive – so much so that it became a new musical medium.

Vienna Symphonic Library and East West sampled entire orchestras, an astonishing feat at the time. Their current orchestra libraries are still very much relevant, as are ones from other developers.

Sampling works especially well for “hit and ring” instruments – piano, percussion, plucked instruments, and so on. There are also sampled choir and other vocal group libraries that are really fun to play. Developers have refined the technology considerably over the years, and anyone who reads Synth and Software knows that there are great sample libraries coming out all the time. You also know that there isn’t a hard line between sampling and synthesis.

So some of my best friends are sample libraries. But as you also undoubtedly know, the technology has limitations.

With sampling, you’re connecting recordings of performances, and making your own expressive performance out of a continuously variable instrument – like the 37 available SWAM instruments –  usually requires a lot of editing after the fact. Transitions between velocity layers on sustained sounds are often audible. Sample libraries have to switch articulations for different sounds – hard and soft attacks, staccato and legato, and so on. 

If the SWAM instruments were sample libraries, they’d have a near-infinite number of articulations, and the transitions would be seamless. Audio Modeling really has solved all the problems inherent to sampling by modeling how the sound changes in response to your performance. Everything happens in real time.

Audio Modeling uses two different sound-generating methods that have a lot in common. One is in their solo winds and string sections, which do use samples, the other is in their solo strings and brass instruments, which don’t.

What the sound generating techniques have in common is their focus on “behavior modeling.”

SWAM stands for Synchronous Waves Acoustic Modeling, and they describe it thus:

“Developed by Audio Modeling using SWAM-W technology SWAM [saxophones in this case] combine innovative performance techniques, physical modeling, behavioral modeling, and multi-vector, phase-synchronous sample morphing to create the most realistic, real-time controllable, expressive virtual… [snip]. [They enable] endless sonic variation and make every live performance unique.”

(But you can still record the MIDI performance in a DAW as with any other instrument, of course, which means it can be repeated, edited, and also that you can use the same MIDI to play other instruments.)

Are you familiar with the Yamaha VL1 synth from 1994? It was the first synth on the market to use digital waveguide synthesis, a technique that uses delays to model sounds. (I’m fortunate to have owned one since then, and it still amazes me every time I play it, but its emulations can’t touch the SWAM instruments’.)

There have been quite a few synths that use digital waveguide synthesis, such as the Korg Prophecy, but what Audio Modeling does with it is highly optimized for the acoustic instruments being emulated. This screen shot is an indiction of the degree of control you have over the instrument model:

There’s also a bird’s eye view of everything (the screenshot below is for the bassoon). It really shows you how much there is lurking behind this interface, should you want to get into it – and you will as you play an instrument and get familiar with it.

So the character of all the SWAM instruments can be tweaked a lot. You can also start with preset models of different styles of instrument (before you create your own).

For example, here’s a comparison of the standard cello and the Bach cello (which could use some EQ and/or ambience adjustments to take care of the resonances):

---

---

Footprint

Even high school kids are old enough to remember when limited computer resources were a constant annoyance with sample libraries. Current computers have made that much less of an issue than it used to be – loading a sampled orchestra hasn’t demanded at least two computers for a few years – but the SWAM instruments make a mockery of our previous travails.

These instruments take up tiny amounts of storage space, in fact there are even modestly limited iOS versions of all but the SWAM string sections (yes, that really means you can play them on an iPhone). A sampled string library might occupy 375GB on the disk, while the SWAM string sections use less than 1GB – much less if you don’t install all the plug-in formats.

But storage is cheap these days; it’s memory access that has been the most limited resource for sample libraries. The SWAM instruments use maybe a third of the RAM that sample libraries require – and it wasn’t long ago that we needed an army of networked computers to load an entire orchestra.

Just for scale, all the string sections from one sampled orchestral library I loaded increased the memory being used by about 8GB for one mic position, and it’s common for them to have three or more positions to mix and match. Each additional mic position doubles the number of voices the computer has to stream, to say nothing of the processing power.

Compare that to about 700MB for all four SWAM string sections. That’s well under 10% of the memory use! And instead of samples from multiple mic positions, SWAM has its own modeled ambience system, called Ambiente, and which we’ll get to shortly.

As to CPU use, this is 50 tracks of SWAM instruments playing in Logic Pro X (Mac Studio M1 Max, 64-sample buffer):

The two test Macs used for this review are that 2022 M1 Mac Studio Max and a new M4 Pro MacBook Pro. I never had to run Logic Pro X at more than a 64-sample buffer (that’s very low) on either Machine, and it was okay at 32 samples when I didn’t have several SWAM instruments going.

The license allows you to install the instruments on up to four computers at a time.

Sound and playing

When you insert a SWAM plug-in and play a note the first time, it tells you why you’re hearing nothing, and that it’s absolutely serious: you want to be using MIDI CCs to control the sounds.

Once you’ve selected a controller template, even if it’s just a standard keyboard, you just play. The SWAM player knows a lot about the controller you’re using, for example EWIs don’t use note-on velocity by default (in fact all but the more recent models just send a fixed velocity).

You can tweak from there very easily, mapping all kinds of instrument parameters using the wiggle and learn method before saving your setup. I’ve been using pedals with the EWI as additional controllers, for example introducing growl to a sax.

Getting totally comfortable playing SWAM instruments does take a minimal amount of practice to refine, but it really is minimal, and you’re still likely to be astonished at the music you’re making right away.

Of course, that does assume you’re a musician (these are instruments, after all). However, it’s hard to imagine anyone finding it seriously challenging to coordinate playing a note and then controlling it with something that sends MIDI CC messages.

Wind and breath controller impressions

Expression pedals, sliders, aftertouch, ribbon controllers… they all work very well with the SWAM instruments, which is to say that you may or may not use a wind or breath controller. So by all means skim through this section if you’re not interested in the impressions of a musician who does. 

The flugelhorn was one of the first SWAM instruments I tried out with my EWI5000, and the obvious real instrument to compare flugelhorn to is Chuck Mangione’s “Feels So Good.” Its second intro (!) a couple of minutes in is a repeated C7sus guitar “comp,” and I started playing the melody along with Mangione when he came in.

But all I could hear was the recording, so I stopped playing to find out why… and the melody stopped too! Obviously, I’d come in early and had only been hearing the SWAM flugel. Yes, it’s that realistic.

Now, there’s one thing about playing SWAM with a breath controller or an EWI that’s unlikely to affect most people: coming from a recorder background, I want to be able to use breath for vibrato. That means controlling the vibrato’s pitch with your breath pressure, not just the volume.

There doesn’t seem to be a way to set that up in the SWAM instruments, in fact the ensemble strings manual advises you not to do that on the controller side (although I’d happily defy the manual if I could figure out a way!). Clearly, Audio Modeling would rather you not commit felonies like playing vibrato on an open string. It wants to create the vibrato for you, at least on the ensemble strings.

However, it took very little time before using the EWI’s bite sensor, which sends a short pitch bump up and down to emulate “reed embouchure” vibrato, became subconscious. That undoubtedly feels totally normal to musicians who play a reed instrument.

Of course, recorder has no particular embouchure, much less a bite sensor. So I found being able to control the vibrato’s pitch and volume separately quite interesting (when playing most of the SWAM instruments I now start with breath volume, and then quickly use the bite sensor to bring in some pitch). And for emulating vibrato some instruments, such as the double reeds and flutes, you want 95% volume modulation.

This is all subtlety that may sound difficult to coordinate, but it really isn’t, because you only vibrato sustained notes and you’re not moving around the instrument.

As virtual instruments go, the SWAM instruments are pretty responsive to double- and multiple-tonguing. The attack response is very quick; it’s the release that’s tricky with wind controllers, which send note offs when your blowing pressure drops below a certain level (and there’s no user setting on the EWI to adjust the level).

To help get around that, the SWAM instruments have a Wind Controller Release Mode that sets the release as short as possible. It does help. And this is not an issue if you play repeated notes on a keyboard, where there’s no ambiguity about when to send note-offs.

Of course, some of the response is a perfectly correct part of the SWAM behavior modeling, so it’s not the same with all the instruments. Similarly, the response to different tonguing attacks varies from instrument to instrument, for example the bassoon – which I’m in love with – responds to different kinds of double-tonguing (“tika-tika,” “toodle-oodle,” etc.).

Realism

Every one of the SWAM solo instruments sounds totally credible to my ears. That even includes the saxes, which are usually a dead giveaway. But here’s the oboe:

---

Sure, every once in a while I’ll hear something that sounds slightly synthy – that is, in a bad way! – to my ears, but more often than not it’s because of my playing. It’s hard to generalize about what goes wrong, but it’s usually the shape of a note in the middle of a phrase.

Now, that’s not to say that it won’t sound synthy if you do clearly silly things like using sax-style pitch vibrato on a bassoon. But it’s not the actual instrument sounds that are off.

Bear in mind that sample libraries also break character every once in a while.

Here’s some bassoon noodling that hopefully doesn’t have any sax-style vibrato:

---

If you’re going to make instrument sections with multiple players (three flutes, etc.), you set up different variations of the same instrument, either by calling up different presets or by tweaking one they provide. There’s also a setting to offset the phase slightly, and you just pick a different number for each instrument.

While you can certainly get away with just layering the same instrument to create a section, especially if you’re playing live, in practice it may take some work to make it sound totally convincing that, for example, there are two flutes playing. Playing each part in multiple times may help, but you’re up against the same issue musicians always face doubling ourselves on acoustic instruments: you agree with yourself 100%, so it can sound like double-tracking rather than two players.

Randomizing in a DAW helps, both for creating sections and with octave unison parts in general. Anything to vary your performance is worth trying, for instance switching expression controllers or maybe playing the occasional part from the keyboard instead of EWI. 

Ironically, it’s the subtle performance details that put the instruments in sharper focus with 1-man SWAM ensembles than with sample libraries.

Ensemble strings

The ensemble strings do sound good, and they really warrant a separate article in the future. As an overview, these are individual string sections, and you select the number of players in each – although subjectively they’re innately small-sounding sections, even with five cellos and a big ambience setting.

As with real instruments, you can only play one section at a time. If you want all four (plus divisi) parts under your fingers while you’re writing or playing pads, you have to use a sample library, then come back and play the SWAM first violins, then second violins, violas, etc. one at a time.

While the SWAM solo strings lend themselves very well to playing with a wind controller (as long as you don’t need double stops), the SWAM string sections are probably best played using a keyboard plus some kind of MIDI expression controller – breath controller, expression pedal, etc. for the volume.

Among the reasons to use a keyboard (or, say, a Linnstrument) for the ensemble strings is that they’re set up to use velocity. Increasing the velocity sharpens the attack, and progressively higher velocities of a second note while holding a first speeds the transition gradually from stinky cheese portamento all the way up to very smooth legato at high velocities. There are keyswitches to choose whether to start with upbows or downbows, with a remarkable selection of choices, such as the bow and finger positions.

In addition to arco, the SWAM strings have harmonics, sordino (mutes), tremolo (different speeds), pizz…  okay, they don’t have the recorded 20th century Penderecki-ish effects in, say, Audiobro LASS – or unique effects such as the ones in Realitone Sunset Strings and Nightfall – but what they can do with more standard performance techniques is pretty remarkable, and it takes very little effort to make them sing.

Whether the SWAM string sections are replacement for a good sampled string library is a subjective call. For me they’re an outstanding complement. Being very critical, subjectively there are times when it requires some suspension of disbelief to hear different string players in each section, rather than clones of the same player. On the other hand, you’d have to work hard not to create an undeniably musical performance.

The ensemble strings – and for that matter solo strings – work well layered with sampled ones for a larger sound. That combines SWAM’s expressive nuances with the sound of a real string section with multiple musicians recorded in an acoustic environment.

Positioning

Audio Modeling includes a room simulator, called Ambiente, with all their instruments. A stand-alone version for other instruments (which we haven’t tried) is also available. It really is a room simulator rather than a reverb, because it positions the instruments in a space and is an integral part of the model sound.

The effect is convincing. If this were a review of a regular reverb unit, it would pass all the usual tests – good small spaces, sound that sticks to the instrument, no annoying sparkles or plastic sound. Flute is very good at revealing that sterile plastic reverb sound, and Ambient has none of it. 

As you can see in the following screenshot, Ambiente talks to all the SWAM instruments you have loaded in your DAW. It will give you a loud red warning if multiple instruments are assigned to the same spot.

Fine-tuning the instrument models is different from adjusting familiar synth parameters, but adjusting Ambiente’s modeled spaces is more intuitive than using the technically-named reverb parameters we’re used to – RT60, predelay, etc.

You choose from about 17 spaces as a starting point – listening rooms, studios, theaters, concert halls, cathedrals, etc. As you adjust their absorption material (wood, concrete, etc.) and size, the room type changes. Positioning instruments in the room is as simple as dragging them around the picture of the space, or you can use sliders for the angle and distance from the mic.

If you want to use your own reverb for a SWAM instrument, you can move it very close to the mic in Ambiente, which effectively removes the reverb tail. The ensemble strings all have to be in the same space – by design. But you can assign solo instruments to up to four separate spaces.

Conclusions

The sheer length of this story alone should be an indication of how substantial the Audio Modeling SWAM instruments are. In the area of emulative synthesis, my opinion is that they’re as significant a contribution as streaming sampling was a quarter of a century ago.

Of course there are always going to be nip and tuck improvements, and it’s not like modeling synthesis is brand new – this is just a very successful application of several techniques. But you have to wonder how much closer it’s possible for acoustic instrument emulation to get.

This won’t be the last story we write about the SWAM instruments. One obvious avenue to explore is how to break their strict adherence to real instruments and transform them into new instruments we haven’t heard before. In a way that’s going full circle to what musicians were doing in the ’70s, only we have far more processors today.

It seems that music technology is moving in two opposite directions: toward 1-finger generation that creates and performs the music for you, and what we have here: highly expressive musical instruments for – ducking and running – musicians, ones willing put in the minimal effort to learn how to make them sing.

Take this review as an enthusiastic rave. The SWAM instruments are truly wonderful.

Prices: The price structure is complicated and it seems to depend on how many instruments are included in a product, so we’ll just give you some examples.

Each solo string instrument is $120; the two clarinets (bass and regular) are $170; a bundle of all the solo winds is $750; a bundle of all the solo instruments is $1400; and the string sections are $500.

More info: Audio Modeling is distributed by Ilio.

---