Please list True Stereo Reverbs

[valhallasound]

I define "true stereo" as anything that doesn't mix to mono before being processed by a reverb network. By this definition, algorithms that have stereo early reflections, but mono input late reverbs, aren't true stereo

I definitely agree with you there, Sean. While experimenting with different reverb designs for Verberate, I was actually surprised to find out how important true stereo processing is also in the late reverberation in order to achieve a sense of space even if other techniques are used to de-correlate the outputs.

So this very simple setup would not produce "True Stereo"?



The ER engine could have any numbers of taps and/or allpass filters and/or something else. The above would create 4 completely different outputs which equals "True Stereo".

If the ER engine has decorrelated outputs for left and right that are being sent to the late reverb, this isn't really summing to mono. Simply delaying the right input from the left by 11 msec or so will decorrelate things to some extent. However, this will also cause some sort of comb filter coloration when summed to mono, and this coloration will be present in all of the late reverb. More complicated L/R input decorrelation will create a more complicated comb filter, but it will still be a comb filter.

Meanwhile, a late reverb network with separate left and right inputs (or N channels of input) will have a potentially HUGE number of comb filter outputs, to the point where no fixed coloration is heard. The left and right inputs will be combining in some manner, whether it is through parallel output taps, or via residual energy hanging out in an FDN or allpass delays, but this can sound much better than a single mix of left and right...

That is, better for stereo miked signals. For mono signals, or for mono signals that are panned in the stereo image, these sort of cancellations won't happen when summed to mono. So a mono input reverb, or a reverb with stereo early reflections and a mono late reverb, will work fine there. In such a case, having different delayed/allpassed signals for left and right channels, that are summed to mono, could add coloration that wasn't in the original signal.

Sean Costello

[Warp69]

I must clearly misunderstand something here, since I can't figuring out if you consider the proposed "algorithm" as "True Stereo" or not?!

I was responding to the following statement :

I define "true stereo" as anything that doesn't mix to mono before being processed by a reverb network.

The proposed "algorithm" DOES mix to mono before being processed by a reverb tank/network, which would indicate that you (and Stian) don't consider the "algorithm" as "True Stereo". This would also indicate that you do not agree on the following statement :

Most people would categorize a reverb as "True Stereo" if the reverb creates 4 different outputs for a stereo input - e.i. left input creates a pair of outputs (left and right) and the same is true for the right input. Those 4 outputs are different compared to each other and are mixed together to form a stardard stereo output (L->L + R->L, L->R + R->R).

Mayby you have another definition of mono (single channel) or maybe equals "reverb network" with the whole reverb algorithm (if this is the case - then we agree, no?) or something completely else, but some clarification would help greatly with the definition of "True Stereo".

[Andrew Souter]

I might be using the term "Dual Mono" incorrectly. IIRC, the SKNote Verbtone has a "dual mono" mode, where the left and right channels are not only processed separately, but also have the same reverb network for each channel (i.e. 100% correlated). I have the capability for independent left and right reverbs in a few of my plugins (Room, ÜberMod), but the left and right reverbs will be decorrelated. Most other reverbs with no crosstalk between left and right will have decorrelation between the channels.

In 2CAudio Verbs we have Cross and Width. They handle these topics. For example:

Cross = 0.0, Width = 0.0 translates to:

"dual mono" mode, where the left and right channels are not only processed separately, but also have the same reverb network for each channel (i.e. 100% correlated).

Cross = 0.0, Width = 1.0 translates to:

the capability for independent left and right reverbs in a few of my plugins (Room, ÜberMod), but the left and right reverbs will be decorrelated. Most other reverbs with no crosstalk between left and right will have decorrelation between the channels.

And we have all varieties of gradations between these two axes, and we have many other possible configurations as well...

The "true stereo" or not question, is kinda basic/101 level in the year 2014 as I said already...

But for purposes of naming conventions, I think the definition is generally as Warp describes. This is how it has been used over the years as I understand. If this is not the definition, then all Lexicon verbs fail to be true stereo, as I understand? Correct? You guys know Lex designs better than me...

However I also agree with Warp that there are more sophisticated/superior ways to do things. In fact, I'd like to believe we have implemented some of them...

[valhallasound]

I must clearly misunderstand something here, since I can't figuring out if you consider the proposed "algorithm" as "True Stereo" or not?!

I was responding to the following statement :

I define "true stereo" as anything that doesn't mix to mono before being processed by a reverb network.

The proposed "algorithm" DOES mix to mono before being processed by a reverb tank/network, which would indicate that you (and Stian) don't consider the "algorithm" as "True Stereo".

Arguably, the algorithm you list is true stereo. It would also have coloration that would be avoided in better "true stereo" algorithms.

This would also indicate that you do not agree on the following statement :

Most people would categorize a reverb as "True Stereo" if the reverb creates 4 different outputs for a stereo input - e.i. left input creates a pair of outputs (left and right) and the same is true for the right input. Those 4 outputs are different compared to each other and are mixed together to form a stardard stereo output (L->L + R->L, L->R + R->R).

I don't think that a "True Stereo" algorithm has to generate opposite channel outputs for a given input. For example, ValhallaRoom, ValhallaÜberMod, and the 2CAudio stuff can control the spread of energy over time between channels (Michael Gerzon first described this trick in 1971, using rotation matrices). If this cross-mixing is set to minimum, you have two independent, decorrelated reverbs. This is still totally stereo. It isn't always the best sound, but it has its place, and it is clearly stereo.

Mayby you have another definition of mono (single channel) or maybe equals "reverb network" with the whole reverb algorithm (if this is the case - then we agree, no?) or something completely else, but some clarification would help greatly with the definition of "True Stereo".

A pedant walks into a bar. Well, it's a restaurant with a bar. Technically it's a brewpub since it has an onsite microbrewery.

Seriously. This is getting to be an overly tweaky conversation about terminology. You know how to create true stereo algorithms, so does Andrew, so does Stian, so do I, and so do a bunch of other folks. Most of the older non-true stereo algorithms were this way due to a lack of computational resources, or physical limitations (i.e. ADCs cost a lot of money in the 1980s). At this point in time, true stereo versus not true stereo is a design choice, instead of some big technical leap.

Sean Costello

[Warp69]

Thank you for your edit of your post - this version is much better. I have to admit I dont understand your hostility.

There was a specific question regarding the term "True Stereo" and I responded :

Sorry for not adding anything to the debate - that's me admitting ignorance, btw, so anyone with a good definition of exactly what TRUE STEREO means in this context?

That's a rather good question. Most people would categorize a reverb as "True Stereo" if the reverb creates 4 different outputs for a stereo input - e.i. left input creates a pair of outputs (left and right) and the same is true for the right input. Those 4 outputs are different compared to each other and are mixed together to form a stardard stereo output (L->L + R->L, L->R + R->R).

I will not categorize the above as 'overly tweaky conversation about terminology' nor do I recall I have questioned anyones ability to create "True Stereo" algorithms?! It is obvious by your latest post, that we don't share the same definition of "True Stereo" - fortunately I did write 'Most people' - so the term is undefined atm or have several meanings (subjective ). As far as I understand you don't exclude L->L, R->R (I call them Dual Mono) reverbs from the term "True Stereo".

It would also have coloration that would be avoided in better "true stereo" algorithms.

There exists several products with the proposed topology (or similar). It's just one of many that do have a mono->stereo reverb tank where the overall algorithm still create 4 different outputs (L->L + R->L, L->R + R->R).

[valhallasound]

It is obvious by your latest post, that we don't share the same definition of "True Stereo" - fortunately I did write 'Most people' - so the term is undefined atm or have several meanings (subjective ).

I think "undefined" is the best way of describing "True Stereo." I've seen people argue that only reverbs that have the ability to have NO crosstalk between left and right channels. Those same people argue that Lexicon reverbs aren't "true stereo." I would argue that they are WRONG, but they insist they are correct. So I think that "True Stereo" is either too vague of a term, or too broad. There are a number of possible definitions for the term.

As far as I understand you don't exclude L->L, R->R (I call them Dual Mono) reverbs from the term "True Stereo".

I don't. I'm not saying that these are GOOD reverbs, but they clearly produce a stereo output from a stereo input, without mixing things to mono. These reverbs will preserve a stereo image better than reverbs that are L->L, L->R, R->R, R->L. They won't adequately emulate a real, physical space, but they are "true stereo" in the broadest definition of the term.

It would also have coloration that would be avoided in better "true stereo" algorithms.

There exists several products with the proposed topology (or similar). It's just one of many that do have a mono->stereo reverb tank where the overall algorithm still create 4 different outputs (L->L + R->L, L->R + R->R).

If the early reflection generator can create a good sounding signal when its outputs are summed to mono, then I could see this working. The problem is that any coloration in the summing is there for good. It becomes the fixed frequency response of the reverb signal. I suppose the same thing is true with the shape/spread in the Lexicon stuff, and other products that feed sparse FIRs into the late reverb, but I've always been suspicious of the coloration of this technique as well (even though I've used it).

I don't think that we will get an agreed upon definition for "true stereo," as there are a number of products for which this is claimed, that implement things differently. If someone wants to invent a different term/buzzword for a given architecture (i.e. the (L->L + R->L, L->R + R->R) one, or (L->L, R->R)), it would probably get support.

EDIT: "Coupled Stereo," for example. This is a good description of reverbs that can be viewed as two parallel reverb networks, where the amount of coupling can be adjusted. This covers ValhallaRoom, B2, ValhallaÜberMod (in its reverb incarnations), and other algorithms influenced by Gerzon 1971. It also is a good description of the Capitol Studios stereo reverb chambers.

"Coupled Stereo" doesn't cover the Lexicon algorithms where the left and right mixing is part of the basic architecture and happens pretty much instantly (i.e. Rich Chamber/Plate, 480L stuff). Some other term would be useful for this type of reverb. Once the terms are established, a list of stereo reverbs could be created, that has exact details for the structures.

Sean Costello

[Andrew Souter]

Generally I agree with both of you Sean and Warp and have the feeling that the conversation has reached a certain level of pedantic semantic hair-spliting that is not overly valuable to pursue further... Safe to say we all know about various types of stereo processing within algo verb design as it is a pretty essential part of the effective design of the products we all market...

EDIT: "Coupled Stereo," for example. This is a good description of reverbs that can be viewed as two parallel reverb networks, where the amount of coupling can be adjusted. This covers ValhallaRoom, B2, ValhallaÜberMod (in its reverb incarnations), and other algorithms influenced by Gerzon 1971. It also is a good description of the Capitol Studios stereo reverb chambers.

again, B2 is capable of this type of behavior, but not only this behavior. It has many configurations, the details of which I prefer not to elucidate further in any great detail...

[Beatworld]

It's just one of many that do have a mono->stereo reverb tank where the overall algorithm still create 4 different outputs (L->L + R->L, L->R + R->R).

If a mono-> stereo reverb with 4 different outputs is placed on a stereo AUX send and a hard panned mono track is sent (in Logic via post panning send) to that AUX send should the resulting reverb be panned the same as the source or remain centered/mono ?

[valhallasound]

It's just one of many that do have a mono->stereo reverb tank where the overall algorithm still create 4 different outputs (L->L + R->L, L->R + R->R).

If a mono-> stereo reverb with 4 different outputs is placed on a stereo AUX send and a hard panned mono track is sent (in Logic via post panning send) to that AUX send should the resulting reverb be panned the same as the source or remain centered/mono ?

Any mono->stereo reverb would remain centered/mono. The mono->stereo reverb tank that Warp69 describes would have panning that DEPENDS on the source position for the first N number of milliseconds. I use the term DEPENDS, because the specifics of the input panning all depend on how the early reflections process things, and what sort of model is used.

It is hard to make generalizations about how a stereo->stereo reverb with 4 outputs should respond to a signal panned to a certain position. A "real" room will have the entire image fill up over time, with the image mixing time dependent on the size and geometry of the room. Something like a Lexicon Rich Chamber algorithm will end up quickly filling the whole image, which is closer to "wide mono" in that the original panning is lost. Algorithms that have variable mixing times for left and right might have most of the energy coming from the original source direction, which may or may not be physically accurate, but can sound good for certain mixes.

Sean Costello

[Beatworld]

Any mono->stereo reverb would remain centered/mono. Sean Costello

Thanks Sean, as I thought.

[Warp69]

It's just one of many that do have a mono->stereo reverb tank where the overall algorithm still create 4 different outputs (L->L + R->L, L->R + R->R).

If a mono-> stereo reverb with 4 different outputs is placed on a stereo AUX send and a hard panned mono track is sent (in Logic via post panning send) to that AUX send should the resulting reverb be panned the same as the source or remain centered/mono ?

If you ask to the proposed algorithm earlier in the thread (stereo in and stereo out, but the actual reverb tank is mono->stereo) then a well crafted algorithm would sound very similar to standard well crafted stereo->stereo.

I have created a 100% wet matlab test with the exact same algorithm as the proposed algorithm.

0-7 sec = hard left input
7-14 sec = hard right input
The rest = centered input

http://www.relab.dk/downloads/sound/Panning.wav

I wouldn't call the result centered nor mono for the panned source.

[Beatworld]

If you ask to the proposed algorithm earlier in the thread (stereo in and stereo out, but the actual reverb tank is mono->stereo) then a well crafted algorithm would sound very similar to standard well crafted stereo->stereo.

I have created a 100% wet matlab test with the exact same algorithm as the proposed algorithm.

0-7 sec = hard left input
7-14 sec = hard right input
The rest = centered input

http://www.relab.dk/downloads/sound/Panning.wav

I wouldn't call the result centered nor mono for the panned source.

I agree.
The key here is the stereo in, rather than mono, isn't it?
I would be interested to hear that test 100% dry.
Thanks for posting the wet clip.

[Warp69]

I was trying to demonstarte that you could create L->L + R->L, L->R + R->R (what some would call "True Stereo") even thou the actual reverb network within the overall algorithm is mono-stereo. The behaviour of the panned source is of course related to the processing done by the ER engine (which is a L->L + R->L, L->R + R->R process) - if the algorithm only had mono input or mixed the inputs together before any processing then we could only achieve the sound from 14 sec to the end of the test no matter the panning of the source.

This is what I was trying to communicate with : http://www.kvraudio.com/forum/viewtopic ... 5#p5627705

http://www.relab.dk/downloads/sound/mondryMonaural.wav

[toothnclaw]

I define "true stereo" as anything that doesn't mix to mono before being processed by a reverb network. By this definition, algorithms that have stereo early reflections, but mono input [for] late reverbs, aren't true stereo.

I make this distinction not due to any physical, real world correlates, so much as the fact that mixing a stereo microphone signal to mono will create weird artifacts. If these artifacts are injected into a reverb, it adds coloration that can't be taken away. If the mixing happens inside the reverb network in some manner, the coloration doesn't happen.

But in the real world early reflections and the dry signal are not mixed to mono before being fed to the late reverb, right? Which means that the scheme posted by Warp69 is not of a true stereo reverb, as the 'reverb' part has a mono input.

So an incoming stereo signal is mixed to mono, ERs are mixed to mono, then all four are fed to the reverb tail. This means that any stereo information is lost for the tail. And then the tail generates a stereo output. THAT's not true stereo, although later on in the schematic several signals are combined to produce the final stereo result.

[Warp69]

The positional information will also be lost in a real environment after x msec of propagation time - at some point you won't be able to perceive the direction of the sound source within the tail, at this point you only hear reflections coming from all directions. This is therefor the point where the mono->stereo reverb networks take over from the ER engine.