Conjugating Reverb
Placing Sound Sources in Real-World Acoustic Environments
by Michael Kunkes

An Audio Ease crew recording an impulse response inside the Cathedral in Rouens, Normandy. Photo courtesy of Audio Ease

For the past couple of decades, synthesized digital reverb has been the standard in the recording industry. Digital Signal Processors (DSP) break down reverb effects into two processes: 1) early reflection, the initial sound created when sound waves bounce off a surface, and 2) the reverb tail, the sounds created by feeding reflections into an engine that calculates and synthesizes millions of reflections, with various amplitude and frequency response variations.

The problem with old-style one-size-fits-all DSPs is that they subject all signal sources, regardless of location, to the same reverb effect, and their standard reverb algorithms lack precision. Emulating the natural reverb created by real-world acoustic spaces is a far more desirable solution, and this process is called convolution.
Put simply, convolution reverb (CR) is a technique that allows engineers to place sound sources in real acoustic environments. Almost any real-world acoustic sound can be captured as what is called an impulse response (IR), and thus reproducible in a convolution reverb, where it can be freely applied to any sound. More specifically, IRs are created when a space is “excited” by using a set of sweep tones of varying lengths and frequencies (or a simple starter pistol) to generate sound throughout the entire audible spectrum, allowing the capturing of the space’s total frequency response. Once an IR is professionally captured, it can be modified in a plug-in by various sets of parameters.

Audio Ease's Peter Bakker.

The math behind convolution is centuries old. It was only in the mid-1990s that central processing unit (CPU) speeds became fast enough for Sony, Yamaha and Australian company Lake DSP to come out with their own hardware boxes. They were expensive, sucked the CPU power that was needed to process the complex algorithms, and had very limited sample libraries. But the ice had been broken.

With the advent of the Apple G4 and other fast processors, the limitations on the development of convolution reverbs by slow CPU speed, high latency, delay and lack of parameter control suddenly became more and more solvable, and the post sound industry exploded with software plug-ins.

“It’s not rocket science,” says Bobby Lombardi, Digidesign’s senior product manager for plug-ins. “You’re just taking two signals and multiplying them in the frequency domain. Where the rocket science comes in is how fast you can do that multiplication, spit it out and have a useful product.”

The first user-efficient convolution reverb plug-in, Altiverb, was shipped in 2001 by Audio Ease, and over the last two years, as dozens of new products have proliferated, sound editors, re-recording mixers, recordists and ADR people have discovered that it’s a lot easier to simply place dialogue, music and effects into a preset space (or one they’ve created themselves), all in real time, than it is to go to a synthetic reverb and spend days tweaking a limited set of parameters to get something that still doesn’t sound quite right.

Digidesign's Bobby Lombardi.

“In many ways, the tools that post-production engineers have had to use were leftovers from the music market,” says Jake Thorne, vice president of product marketing for McDowell Signal Process-ing (McDSP). “But manufacturers realized that post is a huge market with expanding budgets, and they needed to start making custom tools for this market. Some users of convolution reverbs are concerned with the playback quality, some with the vastness of an IR library, and still others with the efficiency of the algorithm. But it’s the confluence of all those things that makes for the best product.”

In the spirit of sampling, following are thumbnail descriptions of some key creators of CR plug-in products:

Audio Ease (www.audioease.com), located in Utrecht, The Netherlands, was founded in 1993 by five software development students at the Utrecht Academy of the Arts, Department of Music Tech-nology. They introduced Altiverb in 2001 as a plug-in for Digital Performer, with a Real Time Audio Suite (RTAS) version arriving in 2002 and Altiverb 5, released in March 2005. “Early versions of Altiverb did not have any controls to influence the impulse response,” says the company’s co-owner Peter Bakker. “Now we have a lot of features that can divide IRs into different types of reflections, as well as other controls that directly influence the IR, such as multi-band decay and a preview mode for early reflections.”

Altiverb 5 screen shot showing IR of the Vienna Konzerthaus, waterfall waveform display and speaker placement.

The most unique feature of Altiverb 5 is the speaker placer, which allows an IR to be modified so that a user can virtually take the speaker that was used to record the IR and move it around on the stage. “We do this by modifying the IR data’s timing and levels, and placing different instruments in different places on the stage at specific angles,” Bakker says. “You can use the ‘waterfall’ waveform display to actually see what’s happening, and adjust the decay levels so you actually see why it does what it does. We felt it was important to let the user know where the equipment was placed during the shoot, and on what equipment it was recorded.”

Altiverb has seen heavy use in features, most extensively on King Kong (2006), as well as Superman Returns (2006), The Lord of the Rings trilogy (2001-03), HBO’s Six Feet Under (2001-05) and others. Altiverb 5 for Windows XP has just been released, and Bakker says that the company is working on a TDM-based version of Altiverb for September release. In addition, 2006 will also see the release of a ProTools 7-optimized RTAS version, as well as Intel Mac compatibility.

Trillium Lane Labs' TL Space, Digidesign's first ProTools HD-based convolution reverb.

The company has invested six years in building its vast IR sample library, and releases at least one new IR each month. According to co-owner Arjen Van der Schoot, “Our recordings already include the major concert halls of cities such as Vienna, Berlin, Sydney and Amsterdam––and a wealth of legendary US recording studios––but also the interiors of over 30 vehicles. With Altiverb being used more and more on Hollywood blockbusters, there will be more planes and helicopters, a set of train interiors and more outdoor acoustics, but also, for music production, a series of Germany’s most prestigious recording studios and a collection of French opera houses and theatres.” Altiverb 5 sells for $595 for Virtual Studio Technology (VST), AudioUnit, (Motu Audio System) MAS and RTAS, and $995 for the Time Division Multiplexer (TDM) version.

Earlier in 2006, Digidesign (www.digidesign.com) acquired Wash-ington State-based Trillium Lane Labs, makers of TL Space, which was introduced in June 2004, the first convolution reverb to run on an HD DSP. “We acquired Trillium Lane because there were a number of products they were working on that were desirable to Digidesign,” says Lombardi. “They brought to the table a ProTools HD-based convolution reverb, which was something unique to the platform.”

Waves Audio's Amir Vinci.

The advantages of a TDM CR plug-in, according to Lombardi, are near-zero throughput latency when using ProTools HD Accel systems, and an extremely small throughput delay. “The minimum delay on an HD Accel DSP is just three samples, and that’s for every single plug-in that’s loaded in,” he says. “In a purely mix situation or sound stage workflow, this is unbeatable.” TL Space is currently available in both native RTAS ($495) and TDM ($995) versions.

Like Altiverb, the IR-1, created by Israel-based Waves Audio Ltd. (www.waves. com) was also one of the first real-time convolution plug-ins for common audio workstations, first introduced in 2002. In its current version, the IR-1 V2 stereo version allows total control of all the usual reverb settings, and adds controls for convolution length, reverb resonance, decorrelation control, direct control, tools for capturing personal IRs, convolution start control, early reflection build-up control, dry gain mode, dynamic preset handling and a host of pre-delay settings. The company also makes the IR-360, a surround convolution reverb in a variety of channel convolutions, with a feature set identical to the IR-1. Available in TDM and native for Windows1000, XP and Mac OS X versions, the IR-1 sells for $800. The IR-L, with just basic controls, goes for $400, and the IR-360 costs $1,200.

Package for Waves Audio Ltd's IR-1 realtime convolution plug-in for common audio workstations.

The Waves double-CD library consists of over 120 IRs of venues large and small, created by a dedicated crew of sound engineers and a mobile IR capture set-up that they take with them. “The IR-1 was revolutionary in the way it allowed you to not just do straightforward convolution, but also to manipulate your IRs for specific needs,” says Amir Vinci, Waves’ senior product manager.

McDowell Signal Processing, Inc. (www.mcdsp.com), founded in 1998 and based in Mountain View, California, introduced its convolution reverb product, Revolver, in 2005.

Revolver comes with a complete set of capture tools, and a huge sample library largely recorded by Thorne, who is a former Digidesign employee, as is company founder Colin McDowell. “Our tools are some of the most efficient in the industry,” Thorne says. “Using off-the-shelf equipment––for example, Gene-lec speakers and Rode NT-4 Stereo microphone––the whole process of creating an impulse response, from the time you first generate your sweep to a preset being used in your session, is about five minutes.”

McDowell Signal Processing's Jake Thorne.

Revolver can either be purchased as a stand-alone for $495, or as part of a bundle. Revolver LE, which does not contain capture tools, is available only in the project studio bundle, also $495. The company has just announced a deal to license algorithms from Denmark-based TC Electronics, adding over 40 IRs from its M5000 reverb.

A low-cost product is Pristine Space 1.6, from Russia-based Voxengo (www.voxengo.com), an 8-channel surround convolution processor which company founder Aleksey Vaneev says offers the kind of control and layout not offered in other units. “The key feature of Pristine Space is its ability to work in up to 8-channel surround, with full control over dry or wet reverb levels on every channel. It’s also possible to load up to eight IRs and assign them freely to any of eight convolution channels,” he says. “That way, it’s possible to combine various impulse responses—even those that are strictly stereo—and create a full eight-channel surround ambience.” Pristine Space, which sells for $120, also integrates seamlessly with Voxengo’s Impulse Modeler ($40), a plug-in for IR design that enables users to instantly hear adjustments in IR design.

McDSP Revolver screenshot showing IR Waveform level representation.

The next hurdle, many agree, is the problem of recording IRs in an outdoor setting, which so far has proven extremely difficult to do with any degree of quality. “It’s just much simpler to achieve the required signal-to-noise ratio within a closed venue,” says Waves’ Vinci. “Outdoors, you cannot count on a quiet and still environment and, in certain situations, you need a much louder excitation element or speaker that would be loud enough and linear enough to serve as the right piece of gear to catch the acoustics of a certain outdoor environment.”

“Outdoors is difficult, because you are usually in a place, such as the woods, that absorbs sound, not reflects it,” adds Lombardi. “How do you capture that? I think we will get there with more dynamic products that aren’t just convolution reverbs, but have dynamic matching EQ processes that can be placed on top of convolved spaces to take us into much more complex space emulations.”

There’s also the philosophical question of what constitutes an outdoor reverb, explains Bakker. “Some recordists will say that it comes from the echo phasing of the air moving around; others say it’s sort of an infinite reflection coming back from the clouds. We’ve taken some IRs in the forest. It wasn’t perfect but it wasn’t bad. But it’s very difficult to make something that will be convincing for everyone.”

Pristine Space eight-channel surround
convolution processor from Russia's Voxengo.

The next generation of IR products will depend, as this one did, on revolutions in CPU speed and power that will enable product designers to demand more from the algorithms of convolution. In addition, makers of convolution products will constantly have to find ways to refresh their libraries on a regular basis and continue expanding the possibilities of their plug-in architectures.

“We might be heading to all sorts of places, like using wave field synthesis in order to get greater variety on a single IR that was recorded a certain way,” posits Vinci. “As the technology evolves, this kind of dynamic convolution will become very practical and viable. We will be able to create whole sets of convolution filters that not only sample a room at a certain point, but also sample an entire path, where the editor can literally walk through an ambience track, rather than through an ambient frame. We’d be convolving a whole track of an acoustic signature, rather just a single frame.”

Impulse responses have created a revolution in preserving for posterity the world’s great acoustical spaces, and as with any form of professionally made recording, impulse response libraries need to be copyrighted and protected as intellectual property, an issue that continues to dog digital creators. “If you are going to take something you consider to be unique to your product and make it something that’s only available to your customers, you have the obligation not to make it freely accessible to the world,” says Lombardi.

“High-end recording rooms around the globe have spent millions creating these perfect recording environments that some people walk into and quickly record an IR sweep without telling anybody, then put it on the Internet as shareware. That’s dangerous, and it’s wrong,” Lombardi warns. “It is important to realize that this kind of recorded content is considered intellectual property and we need to acknowledge, justify and properly license IR captures from these companies who want to make them available. Otherwise we won’t have a studio recording industry.

[ return to top ]