Digital Reverberator Face Off
During the development of the HyperVerb digital reverberator we spent some time analyzing common digital reverberators, as well as natural reverberation. This analysis, in addition to years of research, has helped us design a digital reverberator that, in certain respects, offers a significant improvement in sound quality over previous technologies. The two most important improvements are significantly increased echo density, and virtually no coloration in the late response, all while running in real-time on a modest processor (e.g. 150MHz Pentium or 66MHz 56002). One or both of these deficiencies has plagued most digital reverberators since their initial development 30 plus years ago.
The properties of high quality reverberation are very complicated, but for the purposes of this report we will limit the analysis to aspects that can be easily and objectively measured and that are commonly problematic with most digital reverbs. Of course the most important test is simply listening to the reverberator, though other tests can also be useful. For this reason you can also download impulse response recordings for a number of commercial reverberators (ftp.arboretum.com/pub/demos/reverb/) and listen to them as is or convolve them with actual sounds using a product like Sonic Foundry's Acoustic Modeler. Many of these files have large amounts of hiss and/or quantization noise. This is generally due to the impulse recording process and was not necessarily a deficiency of the reverberator. Since this only is a demonstration, such additional noise should not be too much of a problem. Also, since most reverberators can be made to sound "bad" if you tweak the parameters in a certain way, we tried use a generic hall factory preset for each reverberator in an attempt to make things as fair as possible. But since we are still biased with regard to this whole subject, we welcome you to create your own impulse responses so you can verify our results. If you do so please send the impulse responses our way!
All of the tests performed here are based on examination of each reverberator's impulse response. An impulse response was obtained by sending a one sample wide "click" into the reverberator and recording the output. Frequency domain and time domain plots can then be made from the impulse response. From these plots, it is then very easy to make qualitative and even quantitative comparisons based on echo density, coloration, amplitude envelope and other properties.
Echo density, often referred to as density or diffusion, can be defined as the number of "echoes" occurring per second. For example, an impulse response where the energy is concentrated at a few samples (i.e.. most samples are zero) will have a low echo density, whereas an impulse response with energy spread out in time (i.e. few samples are zero) will have a high echo density. To compare the echo density of various reverberators, plots have been made of the first 200 milliseconds of the of the impulse response. To examine the impulse responses in greater detail they can be downloaded and viewed with an audio editing program.
Coloration defines how the reverberator changes the frequency response of a sound. Frequency domain plots were made using Cool Edit Pro, using a 4096 point FFT and a Blackman-Harris window. Successive FFTs were made starting at roughly 500ms to 1000ms and then averaged to get the overall frequency response of the late reflections. Late reflections are those which occur after roughly the first 100ms. In the case of a "well liked" concert hall or room the frequency response of the late reflections will be relatively smooth and uncolored. In other words, there won't be any sharp peaks or valleys in the frequency response. In the case of a poor sounding digital reverberator, there will likely be lots of coloration. For example the common problem of a "ringy" sound corresponds to a few distinct peaks in the frequency response, where as a resonant or metallic sound quality generally corresponds to a very "rough" frequency response.
For the purposes of this report we will assume that a good digital reverberator can generate reverberation that is perceptually indistinguishable from actual reverberation of "well liked" acoustical space (e.g. a concert hall). For a good point of reference, we will first examine naturally occurring reverberation. Of course other "unnatural" types of reverberation may be useful in some cases. Here is the first 200ms of impulse responses from a chapel and a large hall.
(NOTE: Cyan = Left Channel and Magenta = Right Channel)
Looking at these plots we can observe a number of distinguishing features. Some of these features include:
While these features are generalizations, most naturally occurring reverberation has these properties, especially for rooms and halls with "good acoustics." Now lets examine some common digital reverberators, including a TC Electronic M2000 and a Lexicon 480L.
Sound Forge Reverb:
Pros:
Cons:
Digitech GigaVerb:
Pros:
Cons:
Power Technology DSP-FX Reverb:
Pros:
Cons:
TC Electronic Native Reverb:
Pros:
Cons:
Waves TrueVerb:
Pros:
Cons:
TC Electronic M2000:
Pros:
Cons:
Lexicon 480L:
Pros:
Cons:
Arboretum Systems HyperVerb:
Pros:
Cons:
Of course none of this "scientific" data guarantees that the HyperVerb actually sounds better than any of these reverberators, so we welcome you to give it a serious listen. And we don't claim that the HyperVerb is the ultimate digital reverberator. Nonetheless, we think you will find the level of quality very high. Also, keep in mind that many more advanced features (e.g. parametric EQ, crossover points, and a gate) can be added to the algorithm relatively easily if necessary. Other important aspects of the HyperVerb algorithm is that it can easily be scaled for various processing and memory constraints, and that it can be implemented effectively on either DSPs or general purpose CPUs. If you have any questions or comments regarding the HyperVerb please contact Chris Weare at cweare@arboretum.com.
Impulse Response Files:
| File Name | Manufacturer | Product/Type | Preset |
| CoolEdit-ConcertHall.wav | Syntrillium | Cool Edit Pro | Concert Hall Lite |
| Digitech-GigaVerb.wav | DigiTech | GigaVerb | Studio Strat (GSP2101) |
| DSP-FX MediumHall.wav | Power Technology | DSP/FX | Medium Hall |
| DSP-FX MediumRoom.wav | Power Technology | DSP/FX | Medium Room |
| HyperVerb-MediumHall.wav | Arboretum Systems | HyperVerb | Medium Hall |
| HyperVerb-MediumHall2.wav | Arboretum Systems | HyperVerb | Medium Hall 2 |
| HyperVerb-MediumHall3.wav | Arboretum Systems | HyperVerb | Medium Hall 3 |
| HyperVerb-BrightLargeHall.wav | Arboretum Systems | HyperVerb | none |
| HyperVerb-StateCapitol.wav | Arboretum Systems | HyperVerb | State Capitol |
| Lexicon 480L-SmallStage2.wav | Lexicon | 480L | Small + Stage (???) |
| Sound Forge-RichHall.wav | Sonic Foundry | Sound Forge | Rich Hall |
| TCWorks-NativeReverbDemo.wav | TC Electronic | Native Reverb | (fixed demo preset) |
| TrueVerb-MediumConcert.wav | Waves | TrueVerb | Medium Concert |
| WaveLab-HallReverb.wav | Steinberg | WaveLab | (???) |