Illustration: Jack Desrocher
Picture yourself dining in a large cave. As you enjoy your meal in the spacious stone surroundings, a diner sitting at the opposite end of the cave drops a fork. Rather than simply hearing a single soft ping coming from the direction of the other person, you hear several distinct pings in rapid succession that echo from various directions. Once they begin arriving at intervals of less than 50 ms, your ear can no longer discern discrete echoes. Instead, you hear only a wash of metallic noise that fills the entire area, lasting for about four or five seconds. This wash of closely spaced echoes is called -reverberation.
Because of the cave''s large size and hard walls, the character of the reverb is dramatic. Had the cave been smaller or had its walls been covered in plush velour, you might not have noticed the diner''s mishap, because the reverberation would have been less prominent.
FIG. 1: In a large building such as Laukyrkan, a medieval Swedish church, a sound can take more than 300 ms to reach the back-row pews from the altar.
Photo: Courtesy Laukyrkan
Sound travelling along a straight path at 1,130 feet per second takes a finite amount of time to reach the listener. The farther away the source, the more time it takes to arrive. In a large cathedral (see Fig. 1), the distance from the altar to the last row of pews could be about 350 feet. In that case, there would be a delay of approximately 310 ms -between the time that the minister -begins the sermon and the time a pa-rishioner sitting in the back row -actually hears it.
In a church that size, someone in the back row can barely make out the minister''s words because the minister''s voice seems relatively soft and muffled. That happens in part because amplitude diminishes at a rate inversely proportional to distance. By the time it reaches the back row, the minister''s voice is reduced to 1/100 of its original amplitude, or by about –20 dB. Moreover, higher frequencies (for example, those above 4 kHz) are attenuated more rapidly than low frequencies as distance increases—the lower the frequency, the farther it travels. Those higher frequencies are essential for maintaining clarity and intelligibility, especially of consonants. But the factor that has the greatest impact on the sound a person in the last row hears is the sound coloration imparted by the reverberant characteristics of the church itself.
MIRROR, MIRROR ON THE WALL
FIG. 2: Direct sound travels in a straight path toward the listener, whereas early and late reflections make contact with one or more surfaces before arriving at their destination.
What causes the multitude of closely spaced echoes that form the reverberated portion of a sound? When it''s unobstructed, a fraction of the acoustic energy of every sound travels in a straight path from its source to the listener. That element is called direct sound (see Fig. 2). The remainder of the sound is dispersed in every possible direction, eventually reaching a solid surface, such as a marble wall, a wooden pew, a velvet curtain, a metal domed ceiling, or a person. Part of the acoustic energy is absorbed by the surface. A substantial portion, however, is reflected away from the surface at the same angle in the opposite direction. That part of the sound is called reflected sound.
An important factor in the perception of distance from source to listener is the relative proportion of direct to reflected sound. For example, a church-goer in the front row of pews would hear a greater proportion of direct sound because that sound would be traveling a shorter distance. Because each type of surface material absorbs and reflects its own unique proportion of frequency components, the reflected sound takes on a different tone color before moving toward its next destination. In general, softer materials with more surface area, such as ruffled curtains or people, are more absorbent, especially when it comes to high frequencies, than hard, flat surfaces such as marble or metal, which tend to reflect high and low frequencies with equal efficiency.
Some of the first few echoes, called early reflections, travel directly back to the listener''s ears, whereupon, if the traveling distance is large enough, they are perceived as discrete echoes. People''s perception of the room size is directly affected by the predelay time, which is the amount of time that elapses between the arrival of the direct sound and its first early reflection. The larger the space, the more time it takes for reflections to travel back to the listener. A typical concert hall has about a 10 to 20 ms predelay time, whereas a small room might have a predelay that is shorter than 5 ms. As early reflections encounter and reflect off of other surfaces, producing late reflections, the time between each arriving reflection dimin-ishes, and reverb density increases until it reaches a maximum and then gradu-ally decays into silence.
In small rooms, reverb density tends to increase quickly, whereas in a good concert hall, reverb density takes about 100 ms to reach its maximum level. That figure is directly proportional to the volume of the space.
As early reflections combine in the space with later reflections, varying degrees of phase cancellation and reinforcement among reflected frequencies result. This coloration is most dramatic in spaces with parallel and flat surfaces because reflection delay times are more likely to be arithmetically related, thus causing a more hollow or metallic-sounding reverb decay. In well-designed concert halls, however, the large number and complexity of reflective surfaces result in a less regular and more pleasing reverb tail.
HOW LONG CAN THIS GO ON?
FIG. 3: This graph shows a typical concert hall''s impulse response. Progressively softer reflections arrive at increasingly smaller time intervals in order to produce the perception of reverberation.
Although the weekly sermon always ends at noon sharp, the minister''s parting words continue to float through the church as the various reflections take time to die away. In a large church, it can take many seconds for the last word to reach 1/1,000 of its original amplitude, or –60 dB. That decay time period is called the reverb time and is determined by several factors, such as the number and materials of surfaces, the overall volume of the space, and even the relative humidity (see Fig. 3).
Had this been a holiday mass, there would no doubt have been a much -larger crowd. People provide additional surface area for sound absorption, thus shortening the reverb time. If the hard marble walls were draped with soft, highly absorbent velvet curtains, the reverb time would be further reduced. On a particularly humid day, the reverb time would be even shorter. Again, these damping effects are most pronounced with high frequencies, which decay more rapidly than low frequencies. Conversely, if the church were renovated to increase its length and ceiling height, the reverb time would probably be even longer.
LET''S GET PRACTICAL
Now that I''ve covered the basic concepts that determine the reverb experience, I''ll take a look at how they can be applied toward getting the most out of your favorite hardware or software digital reverb processor.
FIG. 4: A regenerative delay line is used to achieve a series of closely spaced echoes. The delay time is normally adjustable.
Most digital reverb units simulate the reflective characteristics of a space by using regenerative delay lines. A regenerative delay produces a series of echoes by feeding a small portion of its output back into its input (see Fig. 4). Because an individual delay line can be used only to simulate reflections that are a fixed time interval apart, most digital reverb units employ a complex configuration of delay lines that are connected in parallel and in series. Each delay line is adjusted to a different delay time so that the resulting reflections occur at irregular intervals. An initial user-controllable delay is often added to the front end of the chain to allow for control over predelay time. Thus, one requirement for creating a convincing reverb is to create a high degree of complexity and irregularity in the pattern and timing of delays to simulate a multitude of reflective surfaces.
In order to approximate the -frequency--dependent effects of various types of surfaces on reverb time, most processors have a variety of filtering and equalization capabilities. High-frequency damping lets you set the relative rate of reverb decay among high and low frequencies. To switch from solid granite to heavy carpet, adjust your high--frequency damping factor from, for example, 0.3 to 0.9, so that high frequencies are attenuated at a much faster rate. (The specific range of values on different reverb units varies.)
Fortunately for most individuals, many of today''s processors use reverb algorithms that take care of the nitty-gritty details of accounting for the delay and frequency-response characteristics of spaces of various dimensions, shapes, and materials. You can simply dial up a type of space—for example, large hall, medium room, or small church—and then fine-tune the apparent size of the space using conventional room-measurement variables such as room volume, which is normally specified in cubic meters. Or you could brighten or darken the character of the reverb by changing equalization -settings.
Some sophisticated processors, such as the Lexicon 300 series, allow for adjustments to the rate and envelope shape of the buildup of overall reverb density as well as to the density of clusters of early reflections. That gives users control of not only the apparent size of the space but also of the amount of diffuse reflections. In general, because it represents a faster buildup of early--reflection clusters, an extreme diffusion setting may result in somewhat decreased clarity in your reverb sound, especially when processing vocals.
Before the advent of today''s digital signal processors, engineers developed a number of electromechanical devices for simulating reverb. Two devices that are still widely modeled in the digital domain are spring and plate reverbs. Both add unnatural coloration to the reverb signal. A plate-reverb algorithm normally uses a high level of diffusion. A good starting point for working with plate-reverb settings is to use short reverb times (0.5 to 1.5 seconds) and high diffusion for processing percussion sounds and longer times (1.5 to 4.0 seconds) with less diffusion for vocals and other sustaining instruments.
Most reverb processors also let you create a variety of nonlinear reverb -ef-fects. One of the more interesting and widely used drum effects is the gated reverb, which allows you to pre-maturely shut off the natural reverb decay, -either abruptly or by placing an amplitude -envelope on it. Another commonly used nonlinear reverb is the reverse reverb, which simulates the effect of reflections that get louder rather than softer over time.
THE CUTTING EDGE
Thus far, I have discussed the challenges of artificially synthesizing reverb using delay lines. Many limitations of that approach have recently been overcome by the advent of the “sampling reverberator.” Instead of attempting to approximate the great complexity of delay patterns by using a trusty digital reverb processor, you can use a sampling reverberator to capture the exact acoustical footprint of an actual acoustical environment.
The sampling reverberator uses a technique known as convolution to cre-ate a filter model of an actual space''s -impulse response characteristics, which are analogous to its original pattern of frequency-dependent reflections. First, a short impulse or noise burst—for example, a balloon pop—is generated and recorded in the desired space. Then, both the impulse and unreverberated sounds are analyzed. Next, their spectra are multiplied to produce a result identical to what you would get if you re-corded the unreverberated sound in the actual space.
Initially, because of its computational overhead, convolution was only available as a non-real-time software application. Free programs, such as Csound (www.csound.org) and Soundhack (www.soundhack.com), and commercial programs, such as Sonic Foundry''s Acoustic Mirror plug-in (included with Sound Forge 5.0), offer that capability. Only within the past couple of years have real-time hardware devices such as the Sony DRE-S777 and the Yamaha SREV1 became available. More modestly priced real-time software applications include PureVerb, the standalone Windows program from CATT, and the recently released MAS plug-in Altiverb from Audio Ease. For those of you who want to try out this exciting technology, you can download a nice selection of impulse--response recordings from my World Wide Soundspaces Web site at http://orpheus.tamu.edu/fredrics/isrc.html.
Gaining control over the myriad parameters found on today''s sophisticated reverb processors can seem like a daunt-ing task. But with a little understanding of room acoustics and a systematic approach to experimenting with a few parameters at a time, you''ll be well on your way to finding that sweet spot in your reverb system.
Howard Jonathan Fredrics is an Emmy Award–winning composer and assistant professor of music technology at Texas A&M University.