Dry sample sets: Implications and limitations

[HMaier]

Hello all,

the simulation of a given room with only one IR (impulse response) is impossible, due to acoustical theory. A given room has an infinite number of impulse responses (of which adjacent IRs may be similar of course).

Every pipe with its location on the soundboard has its specific impulse response. You'll perhaps get an interesting room simulation and sound, but not the original one with the precise mapping of all pipes.

Especially if dry mono recording technique with close micing of the pipes is applied, you loose the essential sound characteristic of this pipe concerning its location. Dry mono recordings have no more room depth. What you get is a point-like sound source with the chance to "pan" the location virtually. Pipes usually have a dipole radiation characteristic, which even changes its characteristic for high frequencies.

Dry sample sets with artificially added reverberation usually tend to sound like a (perhaps very good) digital organ in my opinion, not more. The organ seems to “loose it’s soul".
The dry technique supports the electronic organ approach, not the sound documentation approach. We have discussed that in earlier postings in detail.

If you work with a dry sample set with artificially added reverberation, the organ has completely to be re-voiced, like the pipe organ builder has to do, when he moves an organ to another room.
You get an organ, which perhaps has similarities to a “xyz” organ, but you can’t no longer speak of this organ as to be the virtual “xyz” historic organ.

Another important aspect is that you normally won’t be allowed to enter the organ case for close micing of really important historic organs.

[Stefanussen]

Hi Prof. Maier,

You're absolutely right about there being an infinite number of impulse responses in any given room. I spoke with Brett about this, and he says that ideally, dry sample sets would have one impulse response per pipe. At this time, that is quite unrealistic. He mentioned that even one impulse response per rank would be a happy compromise between one per pipe and one per channel/organ as we are doing now. But even that is beyond our capability due to processing power limitations.

I do think that dry sample sets have their advantages though. Don't get me wrong, I love playing the Metz and hearing the acoustics captured exactly as they are. Myself and others agree that the sound is hard to beat. I guess one of the most obvious advantage to dry sample sets is that you can place them in any environment. I'd bet that if you put a multi channel HW setup running a high quality, well voiced dry sample set in a room with great acoustics, even the toaster bashers with discerning ears wouldn't have a clue.

Also, dry sample sets allow HW users to build up a library from which they can build their own custom organs provided that the sample set creators allow use of CODM with their sample sets.

I would also like to mention something in regard to historical sample sets. I very much appreciate the idea of using HW to preserve organs of historical interest. I also totally agree with you that if the purpose of a sample set is to preserve historically significant organs, then it just seems like it makes more sense to record them wet.

Now, at the risk of sounding historically insensitive, I will say that the reason I am into Hautpwerk has nothing to do with said historical organs. Until I got Hauptwerk, all I had access to was a 2 manual, 9 rank unified pipe organ. It's no concert organ. I take a 10 minute drive to the Mormon Tabernacle for the daily recitals as often as time allows and I get such a buzz from hearing that world class V/208 Aeolian-Skinner. I don't get to play that instrument or anything close to its class. Hauptwerk allows me to interpret and experience organ music in a totally new way.

Bottom line is that the quality and versatility of the organ is much more important to me than the historical significance. And I know I'm not the only member of these forums that feels that way. And as for dry vs wet sample sets. I'd much rather have a 100 rank organ with the best reverb I can manage than a charming 15 rank, 350 year old organ with 100% accurate acoustics. I've been listening to some recordings of the dry St. George's Casavant with reverb added which users have been kind enough to upload to the concert hall, and I'll say that the sound definitely passes as good enough for me.

[HMaier]

Yes, thanks, I understand your arguments.

That's the fantastic thing with Hauptwerk, that you can have both approaches, the "wet" and the "dry" one.

I only wanted to explain, why OAM normally uses the "wet" approach for the room based sound documents of historic organs and that this is no waste of computer memory and processor power.

[Stefanussen]

Yep, I very much agree that for historical organs, it's pretty hard to beat wet recording since it captures the perfect reverb!

I'm still waiting on audio cards that render sound the way a video card renders a 3D scene. Can you imagine modeling a pipe organ, defining material, wind, etc parameters, and then have the computer generate the sound from scratch? Now THAT would be something! :)

[Csaba Huszty]

Dear Sirs,

Please let me add my humble opinion to the discussion above:

A room is a (1) bounded-input-bounded-output stable, (2) causal, (3) linear, (4) time-invariant (5) multiple-input-multiple-output (MIMO) system In fact, it is an Infinite Input Infinite Output System, but in practice, there is no need to have infinite inputs and outputs, a finite number of them is just enough to have a *really* accurate simulation of room acoustics for auralization purposes. For measurement purposes to extract room acoustic parameters even fewer positions are enough (see ISO 3382 Annex A.4.).

A key factor or convolution-based room simulation is the tail part of the impulse response which has to be post processed, because real-life measurements do not provide the dynamics our perception requires.

Now imagine that you record an organ dry in a reverberant room. You need to get close to the case (if not inside), closer than the so-called critical distance. Within this distance the direct (dry) field is dominant, beyond this distance the wet field is dominant. As we move closer we have more direct sound so the wet sound may be negligible or acceptably low. This distance varies greatly hall-by-hall, so it has to be calculated case-by-case in order to be able to record a sample set 'dry'. But as soon as you approach the organ case, the assumption that the organ is a 'single' (point or whatever) source fails. This results in the fact that mono recording is not enough any more. Also, one can assume that there may be situations where it is not possible to record an organ outside the case with an acceptably big amount of dry sound due to the shortness of the critical distance.

My big doubt about recording an organ inside the organ case is that it is very far from what organist and the audience hears, and it is - at this point of my knowledge - very difficult to foresee how an appropriate simulation is possible even if I assume theoretically an infinite amount of computing resources.

The term you used previously "every pipe has its impulse response" seems to be wrong and needs some clarification for me, because of the following (and please kindly correct me if I misunderstood something):

A single impulse response characterizes the transfer between ONE input and ONE output of an LTI, stable, causal system (such as the room assumed). A single pipe can be such one source, and we can define a single point where the listener sits in the room (a perfect omnidirectional mono microphone), but it is theoretically difficult for me to see how this impulse response can be measured or how it will look like if modeled.

I can imagine two cases for a measurement:

A) we either measure a the pipe itself and compare it to a reference pipe measurement in the lab or
B) we measure 'something' on-site that we post-process 'somehow' to acquire the impulse response corresponding to the characteristics and the location of the pipe.

Either case a pipe impulse response, which should probably be a collection of them for even one single pipe should take into account that:

1) spatial filtering: the pipe is directional, and every pipe is probably slightly different, directed towards different angles. All of these data have to be measured. Note that a 'correct' directional measurement has at least a 1-degree resolution for all angles in 3D and hold phase information as well (so it is complex), assumed that the directional measurement is conducted in the far-field of the pipe. The nearfield should also be monitored.
2) temporal filtering: the pipe represents a narrow-band excitation, but it is theoretically difficult to determine the exact frequencies and bandwidths of its excitation, therefore an impulse response acquired from a wideband sound source (case B) cannot be easily filtered according to a pipe
3) case reflections: the pipe is located somewhere in the case that interacts with the other pipes and where sound reflects, which is again difficult to model.

I can hardly imagine how case B can be realized due to problems of sound source size and theoretical difficulties of designing post processing methods, but I can imagine that it is possible to do case A: measure a single pipe in a laboratory and later in a real organ and calculate the impulse response from these measurements corresponding to each pipe. Directional information is still a problem to be solved, but even if it is omitted, it is rather an unrealistic idea to do this for every single pipe, not just for historical instruments where the pipes are not accessible at all.

Therefore I think the per-pipe impulse response approach is (as far as I believe now), not realizable - even if there is infinite computational power - due to the lack of measurement information.

Please correct me if I see something wrong.

Instead of that, I think it is better to 'narrow' the pipe organ into a finite number of close but outside-the-organ-case microphones to have all the natural spatial and temporal effects happening at the case recorded as-is, and later extend that multi-channel sound into a wet environment in (e.g. 7.1) surround by means of convolution based auralization using real room acoustic measurements that are post-processed adequately. According to the Huygens-theory, one can see that this can be a good approximation if there are enough microphones and there is a well-behaved acoustic environment (having a relatively large critical distance).

In fact it is no secret that this is the idea that we follow in our current and upcoming instruments.

All the best,
Csaba

[B. Milan]

Since this topic came about just after the release of the new dry Skinner set I feel I should make some remarks.

I think we have two different ideas on how the close/dry recorded pipe samples can be/are to be used. If we are assuming only for home/studio use with reverberation added then it is an entirely different situation than having a church installation and using dry samples which are sounding from a far away speaker chamber with multiple audio channels. Let's not forget that there are many people using Hauptwerk in churches and the larger the room the dryer the samples should be since the room will add its own impressions to each pipe sample.

As Hauptwerk moves forward we are spreading the marketplace to not only more home users, but also into churches that are looking for a high quality virtual instrument (much more so than many of the current digital organs available.) With this in mind we chose to record samples inside the case at a very close proximity to each pipe moving down the wind chest as the rank progresses.

To say there is limitations in this then you must accept that there are limitations within the wet sample sets as well since for one, they cannot be used in acoustic spaces very well (or at all as per some sample sets) and two, the tremulants suffer (unless you sample ranks with trem off and on=2x the memory.) These topics and ideas have all been discussed here on the forum previously and I don't want to repeat all that has been said. I think we all know what can and cannot be done with each type of instrument.

We need to realize there are uses for both and there is a larger market than just the home user looking for the natural acoustics. We offer both methods in our variety of virtual instruments and I think our audio demonstrations speak for themselves as well as user feedback.

Thank you.

[bcollins]

If it please the court, I would like to add my 2 cents worth to this conversation. In my humble opinion, I think I can speak from experience from the view point of a builder/installer of a Hauptwerk organ in a large space.

First off, let it be known that I am not by any stretch of my imagination an acoustical expert. So my input is not based on any scientific or academic theories of acoustic modelling. It is only based on real application in only one room - specifically the worship space at Zion Lutheran church.

It has been (only) two years now that I have been a Hauptwerk user. And in those two years, I have spent countless hundreds of hours of research and study just learning about the theoretical design and application of this wonderful instument - the pipe organ. And at this point I am only approaching the level of knowledge that many here have already by virtue of the fact that they are a professional organist.

I am however, probably in the (rare) position of spending additional countless hundreds of hours of hands-on practical application and experimentation in the pursuit to create a virtual pipe organ which sounds as close to the real thing as possible, such that all but the most discerning and trained ear will believe it is a real pipe organ.

*****
The use of wet pipe samples in a large acoustic space i.e. a church or concert hall is known to be undesirable. The conventional wisdom is that “dry” pipe sample should be used because the room in which the virtual pipe organ is installed will have is own acoustics or reverb. The reverb tails of a wet pipe sample can be trimmed to improve the result, but there will still be the recorded reverb of the original instruments room within the attack and speech loop portion of the sample. So the result is mixed at best.

For example, the sampleset of the Silbermann organ at St. Marienkirche by Milan Digital Audio is an extremely good “wet” sampleset. However the organ is in a church that is an extremely reverberant space. So, even when the reverb tails are trimmed the reproduction of the pipe samples into a room that has natural one-second decay produces a very unnatural sound.

I have experimented by using many different pipe samples for the Hauptwerk virtual pipe organ at Zion Lutheran church. Based on this conventional wisdom, I have experimented extensively with “somewhat dry”, “mostly dry”, “dry”, and “very dry” pipe samples.

For the sake of argument let's reference these examples:

Somewhat Dry: FBR Schantz / Jonathan Orwig
Mostly Dry: Daniel Prib (wet) Chamber organ by Sonus Paradisi
Dry: Compenius Organ by Yves Petit-Clerc / Milan Digital Audio
Very Dry: Daniel Prib (dry) Chamber organ by Sonus Paradisi

My experience has led me to believe that 1) which pipe samples are chosen for the successful creation and installation of a virtual pipe organ depends on many factors - such as the natural room acoustics of the space where the organ is installed, and 2) the implementation of the audio reproduction system. Specifically, how the speakers are implemented.

The old adage that the most important stop in a pipe organ is the room / space into which the organ plays is also (especially) true for a virtual pipe organ.

It is my opinion - and this is the essence of my dissertation - that “dry” pipe samples can be too dry under certain circumstances.
For example: when the natural space is already acoustically dry, as many modern churches are, due to excessive carpeting, padded pews, and generally undesirable building materials such as cedar ship-lap siding which almost seems to suck up more sound than it reflects.

In these modern times - in America, at least - a worship space has to accommodate distinctly different styles of service music. Such is the case at Zion where we have two services; a traditional service with organ, piano, choir, and solo (vocal) music – all un-mic-ed and “acoustic” - and a “contemporary” service with piano, guitars, bass, drums, singers, etc. which all use sound reinforcement (microphones and P.A. system).

Were it not for the wall to wall carpeting on the chancel floor where the contemporary worship musicians “perform”, the music would sound muddy due to undesirable reflections.

The result of the acoustic accommodations to support the contemporary worship is a relatively dry acoustic in the worship space. At Zion we have about one-second decay (maybe), and that’s without anybody in the pews.

Let’s examine for a moment a real pipe organ, in a real church with somewhat dry acoustics: Even at the console – which may be 15 to 20 feet from the casework and the façade pipes – there is going to be some natural reverberation. Why? Because most of the pipework is encased, if not enclosed, in a hard wood enclosure.

Even the Pedal and Great - though not enclosed behind swell shades - are still surrounded by a hard wood case that not only directs the sound outward, but produces reflections and natural reverberation. So even though the area where the console sits may be carpeted from wall to wall, there is still some natural acoustic reverberation at the point where the organist hears the organ. And no matter how dry the room is due to carpeting, padding, banners, draperies, and clothed bodies in the pews – there is still going to be some more reverberation by the time the sound gets to the back of the church.

Contrast this to a virtual pipe organ which uses for its pipe sources “very dry” pipe samples reproduced into a large but dry space with speakers – unenclosed. As it turns out, at Zion this sound is almost as unnatural as the clipped wet samples. The result of using samples which are too dry resembles a reed organ or a harmonium. The phenomenon is especially true at higher pitches.

I realize this may be considered blasphemy to some, but the idea here is to create a virtual pipe organ which sounds as close to the real thing as possible, such that all but the most discerning and trained ear will believe it is a real pipe organ. This has always been my goal at Zion, and I have spent two full years now trying to attain this lofty goal.

The importance of using a multi-channel audio system with as many speakers as possible cannot be stressed enough. This is why only MONO samples can be used. But equally important is using the right pipe samples.

I have come to the conclusion that there are two solutions:

The first, and probably the better of the two, would be to build a hard wood case - maybe even complete with metal façade pipes to add additional reflection and resonance - big enough to contain all the speakers used by the organ, and provide some natural reflections and reverberation. This is a practice that I haven’t had the luxury of trying, but I believe is basically the approach taken by Ogletree and Marshall at Trinity Wall Street – where the speakers are encased in the original pipe organ's casework.

The second, which I have had the most success with, is to avoid pipe samples which are too “dry”, but rather have just the right amount of original acoustics in the recordings, and whose reverb tails are clipped at about 150 – 200 milliseconds with a 135 – 185 millisecond fade.

The challenge is in matching the right amount of original recorded acoustics with the amount of natural reverb in the acoustic space.

The Hauptwerk Virtual Pipe Organ system includes a feature whereby pipe samples can be trimmed of their natural decay. And this tool is extremely helpful in improving the reproduction of “somewhat dry” pipe samples into naturally acoustic spaces. But in my opinion, by its very design, it is more useful by the home organist wanting to control their own reverberation with the use of artificial reverb.

Let me explain. In a real pipe organ, speaking into a natural acoustic, the lower pitched pipes will reverberate longer than the upper pitched pipes. And so the system is designed to trim the decay of higher pitched pipe samples “shorter” than the lower pitched pipe samples within a given rank. And in a small space with added artificial reverb, the result will indeed be more realistic.

But think about this for a minute. When amplified into a large, naturally reverberant space, those same lower (8 foot) pitched pipes are going to decay naturally - slower than the higher (2 foot) pitched pipes. I have found through hours of trial and error - in the room at Zion for example - that the inverse of the Hauptwerk logic is necessary for the most realistic replication of a natural pipe organ sound.

And, against what might otherwise be considered conventional wisdom, the best success has come from using "very dry" samples for 16 ft and 8 ft pitches, [less] dry samples at 4 ft and 2-2/3 ft pitches, "mostly dry" samples at 2 ft, and "somewhat dry" samples at 1-1/3 ft to 2/3 ft pitches.

Therefore it became necessary, for the most successful virtual pipe organ installation, to prepare the pipe samples manually - by trimming and shaping the release, and in some cases using Pipetune to either increase or decrease room acoustics within the speech portion of the samples - rank by rank, octave by octave - prior to the loading of the sampleset.

The effect is subtle. For example, for the organ at Zion Lutheran Church, I have pre-trimmed and faded the 16 and 8 foot ranks to a median 150 milliseconds after the release marker. The 4 foot ranks are trimmed to 185 milliseconds and the 2 foot and above ranks are trimmed to anywhere from 200 to 250 milliseconds (2/3’ pitch).

All of this is dependent of course on how “dry” or “wet” the natural acoustic space is. As Brett points out, the larger [actually wetter] the room the dryer the samples should be. I would love to have the luxury of a church with better acoustics (2 plus seconds), in which case I would use more "dry" samples and the reverb tails would be pre-trimmed much shorter. Although the variation may be tighter, for example the range of decay length after the release might be 85 milliseconds to 135 milliseconds.

As you might conclude from my remarks, no given sample set is appropriate at Zion. And from my perspective a sample set is not an instrument, but rather additional inventory of pipe ranks. A virtual pipe organ in a church must be carefully compiled and prepared to fit the room's natural acoustics. All this before voicing, balancing and regulation begins. And I would like to point out that even though the room acoustics are somewhat dry at Zion, absolutely no artificial reverb is added. I rely strickly on samples which have just the right amount of natural room acoustic in the speech and decay.

Therefore, I don't think there is any hard and fast rule or method to producing a "dry" sample set. A variety of "dry" sample sets is desirable; close mic, near mic, stereo, mono, etc. So that all those that desire can experiment in their space, with their acoustics/reverb - natural or otherwise - to acheive the sound they want.

Thank you for your time considering this was more like 2 bucks worth, which still isn't much. ;)

[Stefanussen]

Hi Bob, thanks for your comments, I think there's a lot of valuable information in your post. I don't have the amount of hands-on experience that you've gained by working on the Zion organ, but your experience confirms a suspicion I've had for a while now.

I've suspected that there's no hard and fast rule for dry vs. wet sample sets when it comes to a 'real' virtual organ installation as opposed to a Hauptwerk setup in your living room. I live in Salt Lake City, HQ of the LDS church. Since I'm LDS myself, and an organist, I've played in quite a few of our chapels. These chapels are not built like cathedrals! They are pretty small and have carpeting everywhere. This translates in to almost no reverb. For some of these buildings, a pipe organ would be a disaster!

The new Conference Center across the street from the Tabernacle is massive and can hold 21,000 people. It has a brand new 2003 130/V Schoenstein organ in it; this organ sounds amazing, 18 months of voicing by Jack Bethards went into it. The sounds are smooth as butter, as opposed to a baroque German organ for example. The mixtures give the organ a wonderfully terrifying metallic sheen. And guess what... the organ uses a complex multi-channel audio system to very subtly amplify the sound of the organ to balance the sound when the auditorium is filled to capacity. What really interested me was to learn that the reverb for the conference center organ is digitally added. They must have done this knowing that an organ has to have some reverb to sound its best. If anyone doubts that this can be done effectively, I encourage you to go listen to this organ.

If I were installing an organ into one of the chapels in which I regularly play, I would probably go with a wet sample set since these buildings are almost completely dry. I think the only time to install a totally dry virtual organ is when the room acoustics are such that they provide a pleasing reverberation. And often, this also means being installed in a sound chamber much like a pipe chamber. Bob makes some very good points about the acoustical effects of a pipe chamber.

[toplayer2]

Especially if dry mono recording technique with close micing of the pipes is applied, you loose the essential sound characteristic of this pipe concerning its location. Dry mono recordings have no more room depth. What you get is a point-like sound source with the chance to "pan" the location virtually.

My comments are relevant to a virtual organ in a typical domestic room (small, non-reverberant).

I agree entirely with Csaba Huszty's point:

But as soon as you approach the organ case, the assumption that the organ is a 'single' (point or whatever) source fails. This results in the fact that mono recording is not enough

After many hours comparing mono with stereo versions, I have found that the stereo samples produce a very important spatial effect that is critical to recreating a three dimensional acoustic portrait. Stereo samples come at a cost of doubling the needed computing resources, but the spatial imaging is well worth the price.

Also, in agreement with comments made by Prof. Maier in another thread, using near-field monitors is also quite critical to achieving this effect.

Likewise, convolved reverb is essential to recreate a believable simulation of an acoustic space suitable for the organ. Experimentation has led me to a ten audio channel configuration with six IR microphone positions, two closer in but biased left, another pair close but biased right, and a third pair for far rear. The four close reverbs are mixed with dry signals in four of the eight front channels. Taken together, the resulting sonic image has a tangibly solid "you are there" sound that is very satisfying. The speaker positions for five pairs (A thru E) and two subwoofers are as shown below:

. [SW] . . . . . [D] . . . [A] . . . . . . [SW] .
. . . . . . . [C] . . . . . . . . . . [B] . . . . . . . .
. . . . [B] . . . ..------------. . . . . [C] . . . .
. [A] . . . . . . .| o ------ o | . . . . . . . [D] .
. . . . . . . . . . . . |||||||| . . . . . . . . . . . . .
. . . . . . . . . . . . |||||||| . . . . . . . . . . . . .
.
.
.
.
. . . . . [E] . . . . . . . . . . . . . . [E] . . . . .

Rather than using audio channels to mitigate IM as many users prefer, to my ear there is more benefit to deploying them to enhance imagery and to segregate ranks with similar harmonic content to reduce the "signal mixing" effects so well described by Colin Pykett.

Joe Hardy

[Ken]

Stefanussen wrote:

If I were installing an organ into one of the chapels in which I regularly play, I would probably go with a wet sample set since these buildings are almost completely dry. I think the only time to install a totally dry virtual organ is when the room acoustics are such that they provide a pleasing reverberation. And often, this also means being installed in a sound chamber much like a pipe chamber. Bob makes some very good points about the acoustical effects of a pipe chamber.

The Mormon chapels have been purposely acoustically designed for the spoken word. They are very acoustically dry and have huge speaker systems usually running down the full length of the chapel ceiling. Although the building/sound system is very good at distributing the sound of someone talking into a microphone, unfortunately these buildings are not very suitable for organ or choirs. These chapels are quite a contrast to the Mormon Tabernacle which has a lovely acoustic for organ and choir.

Organs are used in these chapels primarily to accompany congregational singing or to accompany the choir (certainly wouldn't use many of them for an organ recital). For this reason it would be inappropriate to introduce an artificial acoustic for the organ alone. The organ should share a common (or at least a similar) acoustic to that of the singing voices that it accompanies. Unfortunately that acoustic is dead.

You could always try to convince the church authorities to remove the carpeting, acoustic tile, etc. ;-)

Ken

[Eric Sagmuller]

Hi Joe,

I assume you are using Gigipulse for your system?