[samiam]

You know, this is one of the best threads I have seen in a long time.

I, too, have long had an interest in reverberation. I have never been fully satisfied with the old-fashioned allpass filter + delay lines technique of reverberation; I never felt this made completely realistic reverbs, especially for medium sized spaces with a lot of reflections (such as a concrete stairwell).

One idea that comes to mind, after reading this thread, is to see if it is possible, with today's computational power, to use something fancier than an all-pass filter for diffusing the sound, such as using a convoluter with a very short impulse response convolution (mainly as an EQ and as a way to smear transients). However, while computing power has gotten a lot better, we may still not be at the point where it's practical to have a few concoluters, albeit short ones, in the core of a reverb algorithm.

While I have made some very satisfying music using the reverb in my old Roland SRV-2000 on guitar, this reverb was really an effect more than it was a realistic reverb. The SRV-2000 had, for me, a good sound, but it didn't really add color or flavor to the sound it was reverberating; it however had a very static reverb tail that is really nice for "infinite reverb" special effects.

I also have a PCM-90 with the dual-reverb card (in effect, a PCM-91). People have been talking about the "Concert Hall" algorithm from the 224 on this thread; I don't know how much the 90's "Concert Hall" sounds like the 224's Concert Hall, but the truth is that I have never been satisfied with its sound. I feel "Random Hall" sounds a lot better, but is a bit thin sounding for me with too much tendency for resonating standing waves.

The PCM-90's "Ambience" algorithm is excellent for getting a realistic sound for small to medium sized spaces (People who claim the Lexicons can't make a realistic space should listen to the PCM90's "Living Room" preset before passing judgment); however it doesn't work too well if one makes the decay too long; too much of a "looping" sound in the tail; the Impulse Responses I have heard from an old EMT 2XX digital reverb have the same problem.

This is the only realistic algorithm the 90 has; the others ("Random Hall", "Rich Plate", "Concert Hall", and the algorithms in the dual-reveb cart) are definitely "effect" reverbs.

I just downloaded for free off of the internet a program by Alelsey Vaneev called "Voxengo Impulse Modeler" which uses a form of ray tracing to create reverberant washes. What one does is design a room by placing walls with various reflective and absorbent characteristics in a virtual 2-dimensional space; one can also set the absorbent characteristics of air. The program then creates an impulse response convolution that can be used by any IR reverb by simulating the sound bouncing off of the reflective surfaces and being absorbed both by air and by walls to create a very realistic reverberant sound.

No modulations to stop ringing tails is needed (or in the case of some of the PCM-90s and Roland R-880 presets, we keep the ringing tails because that's better than modulation for some sources); the sound is a good deal more realistic since it more closely m odels what sound does when it reflects off of walls in a large hall.

Older versions of the program had problems with tails being really grainy. Newer versions of the program solve this problem by allowing the emitter (the point source that the impulse response emanates from in the virtual space) to emit up to 1,000,000 rays (the default of 40,000 rays is fast, but quite grainy). This, for most settings, results in a very deep, realistic wash being generated. If one tries to create a wash with a very long decay time (like the Quantec "Taj Majal" preset), there will be some grain at the very end of the trail; the way I work around this is to take the IR with the long tail, run the IR through the convolution reverb using itself as the convolution, and using the resulting 30-second long very thick wash as an IR for that kind of special effect.

One can adjust the size and diffusion of the space by adjusting only a single parameter. Decay isn't a single knob; one needs to change the damping characteristics of the air and walls to adjust the decay.

It takes about 20 minutes to an hour on my 2-year-old computer to generate a really good wash, but the result is amazing. I finally have a way of simulating the wash of things like concrete stairways using a simulated digital reverb.

The program is a fully-functional free download; the only thing disabled in the trial version is the ability to save one's own reverb design, and one can only create three impulses before restarting the program. One is free to use impulses created with the trial version for any purposes however, and registration is only $40 (considering I spent some $1500 for my PCM-90 about a decade ago, this is a bargain).

The attached .wav file is an impulse responses using the program's "Hall 2" preset; the only changes I have made is increasing the number of rays emitted up to a full 1,000,000 rays, increasing the length of the reverb generated to 16 seconds (by default, it only generates a 4-second reverb), and making the air, draperies, and walls more reflective. I then took the impulse response, opened it up in an editor, removed a single transient peak in the file, normalized the file (in a manner so it sounds balanced in both speakers), the converted it to a 16-bit 44.1 khz wave.

I'm really amazed one can get a sound that matches or exceeds the sound of the old-school reverbs in a program costing only $40.

I wonder how long until Moore's Law catches up so we get to the point where we can just have real-time ray-tracing reverbs to add to our mix. It's parallelizes very nicely; each ray can be calculated by a separate processor and mixed together at the final stage (or have like 50 DSPs, each one doing 20,000 rays or so in real time).