Dutch developer Audio Ease recently broke new ground with Altiverb, the first real-time reverb plug-in that bases its presets on sampled acoustics. As a MOTU Audio System (MAS) plug-in, the current version works only within MOTU Digital Performer. The Altiverb relies on a process called convolution, in which samples of actual acoustic spaces are used to generate highly realistic reverb effects. (For more about convolution, see the sidebar, “Convolution Evolution.”)
Altiverb uses high-powered math to reproduce acoustic spaces, which demands an upscale CPU stuffed with RAM to handle the number crunching. However, the results are unlike the digital reverbs most people are used to.
In spite of Altiverb's powerful capabilities, the colorful user interface is simple and clean. The Altiverb is designed to emulate a rackmount unit replete with 3-D knobs, buttons, and LED readouts (see Fig. 1).
A single large knob, which controls the Altiverb's reverb time by applying a decay envelope to the impulse response, dominates the left side of the “front panel.” The maximum available reverb (100 percent) is the longest that was recorded in the original acoustic space; the knob scales it down from there. A numeric display beneath the knob shows the reverb time as a percentage of the maximum. You can change the value by rotating the knob with the mouse, or by clicking on the knob's edge — the knob will snap to that position. For precise settings, you can also type a value directly in to the Altiverb's numeric display. Whenever you change the reverb time, Altiverb must engage in some heavy processing, so the knob has a small indicator that glows red while the program is recalculating and green when it's ready to go. That usually takes only a few seconds.
Aside from the Reverb-Decay knob, the Altiverb's other controls for shaping the reverb sound are the Wet, Dry, and Predelay knobs at the bottom of the center section. The Wet and Dry level knobs offer as much as +24 dB of gain, which could come in handy in some situations. The Predelay knob lets you add a delay between the dry and wet sounds, and it can be set to positive or negative values. When positive, the delay is applied to the reverb; when negative, it is applied to the dry sound, which means the dry sound can be moved right into the reverb (during or after the early reflections)if desired. In some situations, that can create a fuller, more accurate stereo image that retains important localization cues and still lets you adjust the wet/dry mix.
Impulse responses are selected from a drop-down menu in the center of the front panel. Audio Ease has presets for more than two dozen spaces, including Amsterdam's Concertgebouw, acoustically one of the world's best concert halls. Other spaces include schoolrooms, a bathroom, and a recording studio control room. At the Audio Ease Web site, you can download new impulse responses, including several created by Altiverb users. You can even create your own impulse responses in any room.
Altiverb also provides several impulse responses derived from different mic placements within each space. The mic placements often yield quite distinct sounds from one another; they are useful when the basic sound of the room is right but the exact reverb quality isn't.
When Altiverb is inserted in a mono channel, it offers mono-, stereo-, and quad-output algorithms. Stereo-channel inserts offer inputs that are stereo or a mono mix of both channels and stereo or quad outputs. It lacks quad-input algorithms, so only left-right panning is possible with Altiverb; a signal can't be panned front to back through the reverb.
If a quad-output algorithm is chosen, a mono or stereo channel can't return all four channels, so a drop-down menu (to the right of the large knob) lets you assign Altiverb's outputs 3 and 4 (surround left and right) to any odd/even pair of buses in Digital Performer. You must create another stereo strip (channel or aux) in Digital Performer and select the appropriate buses as inputs to route the surround channels.
On the right side of the Altiverb's front panel is an informative Monitor display that can show any kind of text or graphics pertaining to the selected impulse response. The display can include photos of the building or space from which the impulse response was derived, layout diagrams of the mic placement within the space, information about the recording or the impulse response (reverb length, sampling rate, and so forth), and onscreen help. Forward and back arrows step through the available screens.
Above the Impulse-Response menu, radio buttons allow you to select one of Altiverb's two operating modes: No-Latency/High-Processor Load or High-Latency/Low-Processor Load. In terms of performance, that's where you must choose between the devil and the deep blue sea. When Altiverb refers to “high processor load,” take it at its word. A first-generation G4/500 MHz Mac was not just brought to its knees trying to run a stereo-to-stereo algorithm in that mode — it was left facedown on the sidewalk. Although the reverb played fine, the user interface became unresponsive for perhaps 30 seconds at a time, returning for only about 10 seconds before freezing again. “Difficult to use” would be understating the situation, but I managed to get some practical work done nonetheless. However, keep in mind that I was using Altiverb on just one channel.
Even on a new G4/800 MHz dual-processor Mac, the CPU load was heavy. The quad-output algorithms (which I did not dare attempt on the G4/500 MHz) were by far the biggest offenders; a single instance of the plug-in regularly brought up an excessive-CPU-throughput error message from Digital Performer until I increased the buffer size. Even that fix was a little tricky, because the error message provides a button to access the voice-buffer setting in Digital Performer's Studio Configuration dialog, which would be convenient if that were the setting that fixed the problem. Unfortunately, it isn't; the Configure Hardware Driver buffer setting is the one that needs to be adjusted.
Remember that the sessions in which I received the error messages were not laden with tracks and plug-ins; on the contrary, they had as few as two or three tracks and no other plug-ins at all. In addition to the error messages, I battled odd behavior, such as Digital Performer consistently slamming the Performance Monitor window's Processor activity indicator all the way into overload until I restarted the Mac. The MIDI timing also went completely down the drain in some instances.
The performance issues were not consistent from one Digital Performer project to the next, but I encountered performance problems in every session that used the quad-output algorithms in No-Latency/High-Processor Load mode. The mono-to-stereo and stereo-to-stereo presets were much less problematic as long as I used no more than one of each at a time.
You should also believe Altiverb when it uses the term high latency. In that mode, which definitely reduces the drag on the CPU, the latency was as much as half a second or so on my G4/800 MHz and considerably more on the G4/500 MHz. Fortunately, Altiverb reports the amount of latency (in samples), so you can offset other tracks or edit the time lag after recording.
With essentially no reverb parameters to edit, Altiverb leaves you wholly dependent on your collection of impulse responses to get a variety of reverb effects from the plug-in. For many, the provided impulse responses will be adequate, but for those who are undaunted by the prospect of gathering their own impulse responses, Audio Ease provides additional software and information (in the manual) to get the job done.
To create an impulse response, haul your recording equipment to the desired location, record the sound of a starter pistol shot within the space, edit the recording, and feed it to IR PreProcessor, the companion program that comes with Altiverb (see Fig. 2). IR PreProcessor then churns out an impulse response that you can use with the plug-in.
Instead of a pistol shot, you can use other sounds for the impulse source, including a sine-wave sweep, which Audio Ease provides as a sound file on the program CD. Once you have your new impulse response, you can add all of the photos, graphics, and text you want to the Monitor display by importing JPEG, GIF, TIFF, or PICT files.
Remember that convolution lets you impose the characteristics of any sampled and analyzed system onto another system. For example, a click through another digital reverb or a sine-wave sweep recorded through a particular mic or piece of tube gear can be used to make an impulse response for processing signals. (That is how mic-modeling systems work.) However, the results may not always be what you want or expect; it depends on how the process is carried out.
SOUND AND FURY
So, how does Altiverb sound? I tried it on voice, drums, vibes, synth, sound effects, and a few other random sources, and frankly, Altiverb is flat-out the densest and smoothest native reverb I've ever heard. Moreover, it out-performed an assortment of hardware reverbs I had on hand for comparison.
I heard a clear qualitative difference between the sound of “traditional” digital reverbs and Altiverb. The traditional units produced reverb that sounded more “with” or “on” the source rather than “part of” it. Altiverb blended seamlessly with the source material in a way that definitely felt natural.
On the other hand, traditional digital reverbs have an unnaturally wide soundstage, which is partly how they make things sound so big. Altiverb's stereo algorithms have a narrower, if more even, soundstage than the reverbs with which I compared it. The quad algorithms have a nice sense of envelopment. Presets larger than small rooms seem to have a pronounced delay on the surrounds, which feels spacious on large-space presets, but it's sometimes a little bothersome on medium-space presets.
Altiverb's wide variety of impulse responses offer a broad selection of acoustic environments, but the plug-in retains a characteristic sound. Part of that is its density and smoothness, but almost all of the presets I listened to sounded a little tubby in the 300 to 400 Hz region. Cutting a couple of decibels in that range cleaned it up easily.
Altiverb represents a significant step forward in desktop audio. Real-time convolution is powerful, and Audio Ease has priced Altiverb on a par with other premium plug-ins. The reverb sound is strikingly good in its naturalness. However, the processor demands are more than I've ever seen in an audio product. Altiverb might be leading the market a bit: on a current top-end Mac, a quad algorithm was barely viable. As successive generations of Mac CPUs hit the streets, the processing demands should become less of an issue.
The user interface is clean and friendly, though a few controls are a bit awkward to manipulate. The 64-page manual is nicely written and well illustrated, and it clearly explains a great deal of arcane information.
Audio Ease plans to release Altiverb in other plug-in formats, and the company is also considering adding a quad-input algorithm. Those developments would be welcome improvements, but Altiverb already has everything it needs to sound great as long as you have the horsepower to do the heavy labor.
Larry the Oprovides services as a musician, an engineer, and a producer, and as a sound designer for his company, Toys in the Attic. He has contributed toEMsince 1986.
Altiverb 1.4 (Mac)
FEATURES4.0EASE OF USE3.0QUALITY OF SOUNDS4.5VALUE4.5RATING PRODUCTS FROM 1 TO 5
PROS: Extremely smooth, realistic reverb. Simple user interface. Includes a good collection of impulse responses. Well-written documentation. Useful for applications beyond reverb.
CONS: Staggering processor demands. Lack of parameters creates dependence on impulse response collection. No quad-input algorithms.
Minimum System Requirements
G4/400; 256 MB RAM; OS 8.6; MOTU Digital Performer; 30 MB RAM allocated to the host program
Digital reverb has always been accomplished with digital-signal-processing (DSP) algorithms (traditionally, networks of allpass and comb filters) that roughly mimic the results of the acoustical reverberation process. Convolution offers another approach, in which one signal is analyzed, and its characteristics are applied to another signal. Although there are numerous applications for convolution in audio — from the useful to the bizarre — reverb is created by applying an impulse response to a signal.
An impulse response is the response of a system, such as a room, to an impulse, such as a shot from a starter pistol, a short sine-wave sweep, or some other audio spike. When you produce an impulse in a room, you can easily hear the room's response in the form of acoustical reverberation. To use convolution to simulate a room's reverb, record an impulse and the resulting response in the room. Then, remove the impulse, leaving only the response (the reverb itself). That is the job of Audio Ease's IR PreProcessor software.
Once you isolate the room's impulse response, you can apply it to each sample of any digital-audio signal; that is the convolution process performed by Altiverb. Each sample acts like an impulse, and the cumulative effect of convolving the impulse response with all those impulses in quick succession re-creates the sound of that signal in that room.
In other words, you produce an audio spike in a nice-sounding room and run the recording through a number cruncher to isolate the impulse response. You can then crunch those numbers with your clarinet concerto to hear the clarinet concerto in that nice-sounding room.
Convolution is complicated and extremely processor intensive, so why go to so much trouble to make reverb with it? Because it produces a sound that is decidedly less artificial than traditional filter networks, and it exhibits exceptional density and smoothness.
Until recently, convolution has been available only as a non-real-time process to produce all sorts of sonic effects, not just reverb. Perhaps the first widely available non-real-time convolver was the Spectrum Multiplication function in Emu's samplers. Several programs now have convolvers, including BIAS Peak, U&I MetaSynth, and Sonic Foundry Sound Forge. In fact, Sound Forge's convolver, called Acoustic Mirror, is designed specifically to produce reverb.
The hoopla surrounding Sony's DRE-S777 sampling reverb unit stems from the fact that it's a real-time convolution engine for generating reverb. Sony does this with what I call “big iron” — high-powered, real-time DSP chips.
Finally, there's Altiverb: the first real-time, native software convolution engine (no dedicated hardware). In a way, though, Altiverb also uses big iron — it does its heavy computing with the Mac G4's Altivec vector processor chip.