Reverb Subculture

[zmix]

Originally Posted by Warp69

That's because the dry signal is mixed with the processed sound. It modifies some early taps so you get a faster attack and more tight image just like you descript your own plate.

??????? What?

A single sample impulse was the excitation signal for these examples... there is no possible way that 'dry signal is mixed with the processed sound" since they are separated by at least 4ms from the impulse to the 1st reflection and 12ms to the second one...also the signal doesn't modify the taps.. very confused by this assertion.

[seancostello]

Semi-off topic: What do people use on the Mac to view waveforms? Any good free choices?

I want to take a look at the .wav files of the reverbs at different Depth settings, but the view in Logic doesn't look that great.

[zmix]

Originally Posted by seancostello

Semi-off topic: What do people use on the Mac to view waveforms?

I would recommend Logic. Double click on the audio file and it will open up in the waveform editor. The cursor position will display in whatever units you desire.

[seancostello]

Originally Posted by zmix

??????? What?

A single sample impulse was the excitation signal for these examples... there is no possible way that 'dry signal is mixed with the processed sound" since they are separated by at least 4ms from the impulse to the 1st reflection and 12ms to the second one...also the signal doesn't modify the taps.. very confused by this assertion.

I think that Warp69 was referring to how Depth modified the early taps in the CD Plate, as opposed to the mixing of dry and wet modifying the early taps.

[zmix]

Originally Posted by seancostello

I think that Warp69 was referring to how Depth modified the early taps in the CD Plate, as opposed to the mixing of dry and wet modifying the early taps.

I guess it's the use of the term "Modifying" that is unclear. The CD plate has a simple set of 6 delays that are added in parallel to the 'reverb' signal. The "DEPTH" slider controls the level of these taps as illustrated in the image I posted a few posts up... The addition of these taps does not affect the 'reverb' signal per se... How and where is this "modification" taking place?

[seancostello]

OK, I just looked at the PCM96 Tap Slope impulses in Logic (thanks for the tip zmix!). Warp69 is right, they DO look identical.

[zmix]

I hope that N.S. will chime in here and straighten this out, but I suspect that the "Tap Slope" parameter is broken...

EDIT:I went through every single algo on the PCM96 and not one of them had a working "Tap Slope" in the way that NS or Lexicon describe it....

[Casey]

Originally Posted by zmix

I hope that N.S. will chime in here and straighten this out, but I suspect that the "Tap Slope" parameter is broken...

EDIT:I went through every single algo on the PCM96 and not one of them had a working "Tap Slope" in the way that NS or Lexicon describe it....

Well do the presets use this parameter at all?

If so the preset developers had to have been listening for something.



-Casey

[zmix]

Originally Posted by Casey

Well do the presets use this parameter at all?

If so the preset developers had to have been listening for something.



-Casey

Ha..! Good point.. To be honest I have NEVER used a reverb preset in my life...(I've sure written a lot of them). I'll have a look-see soon.


EDIT: I found a "Hall" preset that used the Tap Slope at a value of +1 and printed it. Then I adjusted the value to -1 and printed that... behold:

[Casey]

Originally Posted by zmix

behold:

Sure do sound the same.



-Casey

[Warp69]

Originally Posted by zmix

??????? What?

A single sample impulse was the excitation signal for these examples... there is no possible way that 'dry signal is mixed with the processed sound" since they are separated by at least 4ms from the impulse to the 1st reflection and 12ms to the second one...also the signal doesn't modify the taps.. very confused by this assertion.

First of all - you have mixed the dry signal (dirac impulse) with the output of the reverb and it's therefor difficult to hear the actual level change of some of the taps.

If you delete the "single sample impulses" from the files it's quite easy to hear what happens.

Originally Posted by zmix

guess it's the use of the term "Modifying" that is unclear. The CD plate has a simple set of 6 delays that are added in parallel to the 'reverb' signal. The "DEPTH" slider controls the level of these taps as illustrated in the image I posted a few posts up... The addition of these taps does not affect the 'reverb' signal per se... How and where is this "modification" taking place?

I would gategorize a level change as a modification.

The 6 early taps does affect the overall reverb sound - the behavior of the Depth parameters on the Plate is similar (yet quite simple) to the attack/shape on other algorithms from 224XL, 480L etc. With higher values of Depth you'll get a fast attack and possible an improvement of the early imaging, just like what Sean described (the Attack control of EOS).

[zmix]

Thanks for the clarification Warp..!

I would define a 'modification' as something which has a causal effect on something else, but certainly perceptual differences are an important part of this discussion. The perceptual masking of the changes in the ER levels by the presence of the initial impulse in the 224 Constant Density plate example above is certainly a valid observation.

[Warp69]

Was that directed at me? I don't quite follow you.

[zmix]

Originally Posted by Warp69

Was that directed at me? I don't quite follow you.

I was thanking you for clarifying what you meant and complimenting you on your perceptual observations.

I think that perceptual skills are as important if not more important than theory in the development stages of reverb design.

[seancostello]

Originally Posted by zmix

I think that perceptual skills are as important if not more important than theory in the development stages of reverb design.

I agree with this. I also think that "golden ears" are useful, in terms of having many skilled listeners critiquing your algorithms. Having the little $49 plugin I worked on going head to head against the big hitters was nerve-wracking, but it was great to hear the feedback. I love reading complementary feedback about my work (who doesn't like hearing nice things about what they do?), but the more pointed critiques and discussions are what will lead to progress in the future.

[zmix]

I think that it is important to understand that the context in which an artificial reverberator is used is just as important to this discussion as is the ability that the reverb has to create a context (temporal, spacial and gestural) for the source signal, and there is a need to distinguish these as two separate arguments when exploring this broad topic. Many great reverbs (EMT250, Lexicon 224) might sound 'bad' to a reverb designer today when auditioned without a source signal but satisfy the contextual criteria outlined above better than many more 'sophisticated' reverbs.

"Le mieux est l'ennemi du bien."

[Warp69]

Originally Posted by seancostello

I also think that "golden ears" are useful, in terms of having many skilled listeners critiquing your algorithms, but the more pointed critiques and discussions are what will lead to progress in the future.

I agree, but unfortunately not many people have "golden ears" and are able to describe what they hear (or want to hear).

Originally Posted by zmix

Many great reverbs (EMT250, Lexicon 224) might sound 'bad' to a reverb designer today when auditioned without a source signal but satisfy the contextual criteria outlined above better than many more 'sophisticated' reverbs

It sounds like you have a preference for obvious chorus like modulation (using linear interpolation) instead of other types of modulation found in 480L, PCM96 and Bricasti (Version 1).

[dale116dot7]

Staring at schematics for a while... the 224 has a 6-bit multiplier but a binary point of 5 so for the purposes of interpolation, it's a 5-bit gain. Updates to the WCS are slow because it used 250ns RAM which is only slightly faster than the 224's instruction execution time. The micro (an 8080) can update a coefficient only during a NOP DSP instruction without thrashing DSP code fetches. The PCM70 improved on this by having an active and an inactive WCS, and there's a signal called 'COPY' which copies the inactive one over the active one during one WC period - a very clever couple of logic gates. It allows less grainy (faster) updates to coefficients - and a 6-voice chorus. WCS=writable control storage (DSP code).

[zmix]

Originally Posted by Warp69

It sounds like you have a preference for obvious chorus like modulation (using linear interpolation) instead of other types of modulation found in 480L, PCM96 and Bricasti (Version 1).

This isn't what I am saying here. I am simply pointing out that in many cases, as in the EMT250 and the 224 for example, there were obvious limitations' that are easily overcome today but the perceptual impact isn't completely understood, hence the "better" becomes the enemy of the "good".

Could you please be a bit more specific about how you define the modulation 'types' instead of grouping the 480 , pcm96 and m7? The pcm96 alone has at least three different modulation types.

Also, I think it's a gross simplification to state that I prefer 'obvious chorus like modulation" because in fact I detest the 'obvious chorus type modulation' in artificial reverb. The paper I cited by David Griesinger is a very good example of how a perceptual goal can be used in a reverb design, particularly when exploring modulation for realism's sake.

[Warp69]

Originally Posted by zmix

Could you please be a bit more specific about how you define the modulation 'types' instead of grouping the 480 , pcm96 and m7? The pcm96 alone has at least three different modulation types.

#1 Interpolation (liniear, allpass, lagrange etc) of delay lines by sine, tri, ran - obvious chorus effect
#2 Crossfading between taps (which probably is close related to both #1 and #3, depending how you define interpolation and coefficients manipulation)
#3 Coefficients manipulation - ex. change the values in the FDN matrix

Originally Posted by zmix

, I think it's a gross simplification to state that I prefer 'obvious chorus like modulation" because in fact I detest the 'obvious chorus type modulation' in artificial reverb. The paper I cited by David Griesinger is a very good example of how a perceptual goal can be used in a reverb design, particularly when exploring modulation for realism's sake.

I would say that obvious chorus like modulation is present on the concert hall algorithms from the 224(XL) and the EMT250 algorithm - it's still a chorus like modulation if you use a pseudo random generator for the modulation. It can be implemented differently - good or bad. I said that it sounded like you have a preference for this kind modulation, since both the EMT250 and the Concert Hall algorithm are in your top 3 reverbs and they have an obvious chorus effect. And it seems that you're most interested in discussing the behaviour of those two algorithms

[seancostello]

Originally Posted by Warp69

#1 Interpolation (liniear, allpass, lagrange etc) of delay lines by sine, tri, ran - obvious chorus effect

This seems closest to how nature works, as the microvariations of the speed of sound that can be traced to temperature variation in an acoustic space will result in time/pitch changes. Of course, a real acoustic space will have variations that are closest to very low frequency random generators (0.2 Hz was cited by Blesser). In addition, each reflection will have independent modulation of varying depth, so the effect will be far more subtle than just sticking a few LFOs in an algorithm.

#2 Crossfading between taps (which probably is close related to both #1 and #3, depending how you define interpolation and coefficients manipulation)

I presume you are referring to non-contiguous taps here. Yamaha has a patent on this, although it sounds like this technique was in use by another big company considerably before Yamaha filed their patent. I need to explore this technique in more detail. For stability reasons, I would presume that this is feed forward only - putting such a crossfader in a feedback loop can rapidly lead to instabilities.

#3 Coefficients manipulation - ex. change the values in the FDN matrix

Since any delay based reverb can be (hypothetically) described as an FDN, this can describe a number of techniques. Most of the ones I have tried don't sound that great.

[zmix]

Originally Posted by Warp69

#1 Interpolation (liniear, allpass, lagrange etc) of delay lines by sine, tri, ran - obvious chorus effect
#2 Crossfading between taps (which probably is close related to both #1 and #3, depending how you define interpolation and coefficients manipulation)
#3 Coefficients manipulation - ex. change the values in the FDN matrix

Well obviously the EMT250 does a lot of both #1 and #3, it's remained highly useful despite (but probably *because of*) it's rather low mode density. The Ursa Major Space station uses a form of #1 and something like #3 though it hasn't really got a similar structure to other reverberators. The 224 uses these methods as well, but it's quite easy remove all modulation in that machine and still have a very useful reverb.

When I listen to the "Random Hall" algo (which seems to use #2 above) or the later Lex algos with the "Spin" and "Wander' parameters I hear a lack of 'distance' in the tail. The sound seems to float like a cloud rather than travel away from the listener. I think the Griesinger paper provides a compelling argument against this sort of modulation for realistic spaces.

How would you characterize the differences in depth that you have observed when experimenting with these three methods?

[seancostello]

Originally Posted by zmix

Well obviously the EMT250 does a lot of both #1 and #3, it's remained highly useful despite (but probably *because of*) it's rather low mode density. The Ursa Major Space station uses a form of #1 and something like #3 though it hasn't really got a similar structure to other reverberators. The 224 uses these methods as well, but it's quite easy remove all modulation in that machine and still have a very useful reverb.

I would describe all 3 of these algorithms as relying on #1 only. [EDIT] I presume that the EMT250 has linear interpolated delay lines in its feedback loops, but I may be mistaken. [/EDIT] Method #3 implies changing the gains of feedback operators, presumably in a unitary fashion. A few examples:

- Parallel delay lines coupled by a feedback matrix, where the matrix is changing values in a unitary method. Blesser has this in his recent patent, but it looks like he removed the claims that cover this particular technique. I had worked with similar ideas around the same time as Blesser, but wasn't too happy with the results. It sounds like cycling through different fixed resonances, instead of smoothly varying the resonant frequencies.

- Changing the coefficients of allpasses embedded within a delay loop. This can result in some nasty sounds, as well as very obvious changes in reverb echo density. When I have tried it, it sounds either like water sploshing in a metal pan (for slow variation speeds), or like tortured souls screaming from the pit of Hell (for faster variation speeds). I used one of my failed experiments as a sound to scare trick or treaters one year.

When I listen to the "Random Hall" algo (which seems to use #2 above) or the later Lex algos with the "Spin" and "Wander' parameters I hear a lack of 'distance' in the tail. The sound seems to float like a cloud rather than travel away from the listener. I think the Griesinger paper provides a compelling argument against this sort of modulation for realistic spaces.

It seems like you prefer the sidebands to build up over time, instead of staying at a fixed level. Method #2, or any other feedforward method (which could be any of the 3 methods, if you change #3 to include feedforward matrices) will probably be better for preserving the pitch of sustained signals. Having said that, I love the sound of time variation in the feedback loops.

How would you characterize the differences in depth that you have observed when experimenting with these three methods?

I need to revisit some of the other methods. Most of my experimentation with #3 happened in the early 2000's, when I wasn't as good at setting the delay lengths of a reverb. I have barely tried #2. Method #1 has so many variations, that I have spent much of my time exploring different variations - and I still feel like I have barely scratched the surface.

[Michael Carnes]

OK guys, finally had a chance to look and to my surprise there's an error in the way Tap Slope is computed. While it is still doing stuff, the effect is considerably more subtle than I'd intended. The basic fix is pretty easy from a DSP point of view, but there are a few consequences. As a result of implementing the parameter correctly, the reverb level is slightly changed. I have some calibration handles that allow me to correct the level to within a dB or so of the existing presets, at least for a nominal setting of zero. For positions away from nominal, there may be some noticeable change, no matter what I do. We'll probably put a tech on individual presets to try to tighten them up, but there will still be some minor differences after all that.

This will not be in the release that's currently in Beta. That's nearly ready to go and it fixes streaming issues for a lot of people. No point in slowing it down. This fix will be in a subsequent release.

Sorry I missed this one. The only thing I can say in my own defense is that I had a number of other irresistibly urgent things on my plate at the time I did presets. With a little more spare brainpower, I'd have caught it.

N.S.

[Casey]

Originally Posted by Nobody Special

Sorry I missed this one. The only thing I can say in my own defense is that I had a number of other irresistibly urgent things on my plate at the time I did presets.

For those of us who know NS; We will all attest on his behalf that "urgent" understates the issues he had on his plate at the time.

It's all good!



-Casey

[zmix]

Well, I am glad that I took the time to write Lexicon about the Tap Slope issue...

[seancostello]

Originally Posted by Nobody Special

Sorry I missed this one. The only thing I can say in my own defense is that I had a number of other irresistibly urgent things on my plate at the time I did presets. With a little more spare brainpower, I'd have caught it.

Well, I am thankful that I NEVER have any bugs in my stuff.

(Note: do not look at the source code of any of my Csound unit generators from 1999. Especially, when I got sine and cosine mixed up in my Hilbert filter. Also, you may want to avoid any of my code in VisualAudio. Or the code I am currently working on...)

[Michael Carnes]

Originally Posted by seancostello

Well, I am thankful that I NEVER have any bugs in my stuff.

I used to be perfect. Right up until the first time I screwed something up.

[dale116dot7]

I've seen bugs in disassembled code that was produced in fairly large quantities. The 1994 GM 6.5L turbo diesel pickup truck had a diagnostic trouble code that would not set reliably. The orignal source code must have been missing a # mark because the code did not make sense, but made perfect sense if you added it in.

[zmix]

Originally Posted by dale116dot7

I've seen bugs in disassembled code that was produced in fairly large quantities. The 1994 GM 6.5L turbo diesel pickup truck had a diagnostic trouble code that would not set reliably. The orignal source code must have been missing a # mark because the code did not make sense, but made perfect sense if you added it in.

And, boy do those diesels sound terrible...

There was a major error in a very popular arcade game and later continued into the home video game market: "Donkey Kong" sold millions of copies in it's various forms and was licenced by over 50 different companies. The game involves navigating an Ape arround a lattice structured resembling monkey bars..

The error?

The game was supposed to be titled "Monkey Kong" but a programmer got *one letter* wrong and it became "Donkey Kong"