Opinion: Wet vs. Dry Organ Reproduction

My intent with this article is not to denigrate wet sampling techniques. The point of this theoretical discussion is to ask, and one hopes answer, the question "Can comparable results be obtained with dry samples and convolution?"

First A Word About Signal Mixing

As Pykett has very convincingly argued*, virtual organs suffer from a phenomenon he describes as "signal mixing". This effect occurs when multiple signals are layered electronically and reproduced through a limited number of audio channels. In practice, a few ranks played together, especially if harmonically dissimilar, can sound reasonably acceptable when reproduced through headphones or a single pair of good quality speakers well positioned in a treated room. But, as more ranks are added, the sound gets harsher and ever more irritating. The solution is to segregate ranks (especially those with similar overtone structures) into separate audio channels. Subjectively, each doubling in the number of audio channels (all other things being equal) brings a comparable amount of sonic improvement. Obviously, the economic and space costs soon add up, so every user will need to decide where the point of diminishing returns sets in.

It has been posited that wet sampled organs do not suffer from signal mixing effects nor benefit from multiple audio channels. If correctly understood, I respectfully cannot accept this line of reasoning. True, reverberation can to an extent mask signal mixing and other defects. [The late Tom Hazelton quipped that reverberation "hides a multitude of sins".] Strip away the reverb tails from any wet sampled organ, and you still have individual ranks that when mixed electrically will progressively sound harsher, IMO.

How To Create A Convincing Virtual Room Using Convolution

It is true that wet samples can have convincing reverberation in part because each pipe is a unique generator occupying a unique spacial position. Can a dry sampled organ and convolver come close to this? I think they can. It does not seem necessary to have one IR instance per pipe to create a comparably complex sonic field. One key is not to mix the channels electrically into an Aux bus to drive the convolver. Instead, pick up the sound from a multi-channel audio system using a pair of good quality microphones. The multiple angles of incidence from the speakers, along with the comb filtering effects, will provide most of the randomization observed with wet sampling techniques. The more instances of convolved IRs, each with a unique microphone position when recorded, the more convincing the simulation. It matters that the reverberation signals emanate from as many unique channels as can be reasonably accommodated so that the sound field seems homogeneous and diffuse. It is very important that the listening room be treated to minimize its own signature, otherwise your brain will believe that something just isn't right (whether using wet samples or convolution).

An advantage to convolved reverberation compared with wet samples is that there is much more possibility to adjust the sound to suit taste and listening room acoustics.

Ralph Glasgal's findings are highly recommended reading:

http://www.ambiophonics.org/

Joe Hardy

* Pykett, "The End Of The Pipe Organ"; "Signal Mixing": http://www.pykett.org.uk/EndOfPipeOrgan.htm#Mixing

[mdyde]

Hello Joe,

Sorry - I respectfully have to disagree.

As Pykett has very convincingly argued*, virtual organs suffer from a phenomenon he describes as "signal mixing". This effect occurs when multiple signals are layered electronically and reproduced through a limited number of audio channels. In practice, a few ranks played together, especially if harmonically dissimilar, can sound reasonably acceptable when reproduced through headphones or a single pair of good quality speakers well positioned in a treated room. But, as more ranks are added, the sound gets harsher and ever more irritating. The solution is to segregate ranks (especially those with similar overtone structures) into separate audio channels. Subjectively, each doubling in the number of audio channels (all other things being equal) brings a comparable amount of sonic improvement. Obviously, the economic and space costs soon add up, so every user will need to decide where the point of diminishing returns sets in.

It has been posited that wet sampled organs do not suffer from signal mixing effects nor benefit from multiple audio channels. If correctly understood, I respectfully cannot accept this line of reasoning. True, reverberation can to an extent mask signal mixing and other defects. [The late Tom Hazelton quipped that reverberation "hides a multitude of sins".] Strip away the reverb tails from any wet sampled organ, and you still have individual ranks that when mixed electrically will progressively sound harsher, IMO.

I believe that, in general, Colin Pykett's 'signal mixing' effect *doesn't* apply to wet sample sets. For example:

Sample say ten notes separately on an organ 'wet' using a binaural head (or similar techniques to preserve room reflections as accurately as possible), and then, with the mics kept in exactly the same positions, record those ten notes played simultaneously.

Now listen to both recordings on headphones (or, almost as good, a single pair of good-quality nearfield monitors in a dry listening environment) so that the listening environment has little or no effect on the sound.

If properly recorded, and disregarding any non-linearities in the original organ (which are fairly negligible in a room acoustic or organ, with the possible exception of acoustic coupling such as 'pitch pulling'), and any non-linearities in the speakers/headphones (e.g. intermodulation distortion) the result should be *exactly the same*.

If particular frequencies add or cancel in one recording then they will do exactly the same in the other because the frequencies present and their relative phases will be the same.

I.e. the result of the 'signal mixing' will be absolutely identical, whether it occurred in air or electrically.

Why? Because a room acoustic is (almost perfectly) a (multi-input - each pipe is a separate input signal/channel) 'linear time-invariant system' (LTI). (If it wasn't then impulse responses and convolution wouldn't be applicable at all for acoustics/reverb.) This means that it's mathematically justifiable to add the (wet) signals together after the room acoustic 'system' has been applied to its inputs (the dry pipes).

Likewise, because the system is linear (LTI), it would be mathematically justifiable (assuming it was possible to capture perfect per-pipe impulse responses and dry pipe recordings) to apply a per-pipe impulse to each dry pipe and then mix (=add) the results together.

For an LTI system (such as a room acoustic) the results won't just be similar - they will be *identical*. That's a direct consequence of the axioms of a linear time-invariant system.

That specific property of an LTI system is termed 'superposition'. For example, it's covered on page 63 in the second edition of Haykin and Van Veen's 'Signals and Systems':

http://www.amazon.co.uk/Signals-Systems-Simon-Haykin/dp/0471378518/ref=sr_1_4?ie=UTF8&s=books&qid=1269257518&sr=8-4

Here's a brief excerpt from it (hopefully it's authors won't mind too much):

Consider the room acoustic as an LTI system that takes an n-dimensional input signal vector x(t), where n is the number of pipes, and produces a stereo output signal y(t) (what the binaural head recorded).

(Or, equivalently, consider the input to the system to be an n-channel audio stream, with each channel being the dry signal from one of the organ's physical pipes, and the LTI system (=room acoustic) to be represented by a 2n-channel WAV file for the impulse response, with two channels per pipe (since the room acoustic maps each input channel/pipe separately to each of the two microphones).)

See also:

http://en.wikipedia.org/wiki/LTI_system_theory

However, it *isn't* mathematically justifiable to add individual dry pipe signals together and then apply reverb to the mixed signal - the result will be different from what you would hear from the real organ in its real acoustic - frequency components and phases will add and cancel differently.

That's precisely what I believe Dr. Pykett's 'signal mixing' effect is describing. He's implicitly talking about digital organs (which use dry samples) and the fact that the result of mixing those (dry) signals together then applying reverb (or a a real room acoustic) to the electrically-mixed composite gives a different result to applying the reverb/acoustic to each pipe individually before mixing (in that case).

That doesn't apply for properly-recorded wet samples because the room acoustic has effectively been applied *before* they were recorded, and thus before they were mixed together. I.e. the relevant 'in air' process has already happened; before the sound reached the microphones.

Rather than discussing again what Dr. Pykett might or might not have meant or assumed in his article, perhaps you could see if Dr. Pykett himself could spare a few moments to respond here?

(Side note: multi-channel audio can still have its uses for wet sample sets in a dry listening environment, e.g.to help reduce speaker non-linearities.)

How To Create A Convincing Virtual Room Using Convolution

It is true that wet samples can have convincing reverberation in part because each pipe is a unique generator occupying a unique spacial position. Can a dry sampled organ and convolver come close to this? I think they can. It does not seem necessary to have one IR instance per pipe to create a comparably complex sonic field. One key is not to mix the channels electrically into an Aux bus to drive the convolver. Instead, pick up the sound from a multi-channel audio system using a pair of good quality microphones. The multiple angles of incidence from the speakers, along with the comb filtering effects, will provide most of the randomization observed with wet sampling techniques.

What you would effectively be doing is using the listening room's acoustic effectively to an extra impulse response per speaker.

But because you're still feeding a mixed dry signal to the speakers, and are using less than one speaker/impulse response per pipe, Dr. Pykett's 'signal mixing effect' would still apply.

So from an acoustic point of view it's significantly better than mixing all of the dry signals together into a single speaker/convolver, and with your proposed speaker set-up the more speakers the better (and the wetter the listening room the better, so that the patterns of early reflections vary as much as possible), but I believe Dr. Pykett's effect will still apply unless you either:

1. Use one speaker per pipe, or:
2. Use a separate impulse response and convolver per pipe, or:
3. Use wet sampling and a dry listening environment.

An advantage to convolved reverberation compared with wet samples is that there is much more possibility to adjust the sound to suit taste and listening room acoustics.

Of course, dry samples have lots of other advantages and disadvantages of their own, aside from any purely acoustical considerations.

http://forum.hauptwerk.com/viewtopic.php?f=6&t=5920&start=180#p45099
http://forum.hauptwerk.com/viewtopic.php?f=6&t=5920&start=195#p45152
http://forum.hauptwerk.com/viewtopic.php?f=6&t=2293
http://forum.hauptwerk.com/viewtopic.php?f=4&t=6251
http://forum.hauptwerk.com/viewtopic.php?f=1&t=464

I know you opted for making your forthcoming sample set dry, and I can fully appreciate the reasons that you might prefer that; it seems generally to be preferred by most theatre organists for various very good reasons (e.g. ability to combine ranks from different real organs, ability to apply and adjust your own reverb, suitability for use in reverberant listening environments, and perhaps the buildings the real theatre organs are installed in sometimes not being acoustically optimal for music).

Aside from the theoretical aspects, I believe the listening room as well as the speakers (and amplifiers) are a vital entity in this whole process. (I'm sure Daniel Cook can verify this first hand). Why I bring this obvious point up is that many users (at least several that I have had discussions with) of Hauptwerk seem to overlook this last, and I feel most important issue(s).

Having a good computer, a good sample set and of course H/W is only on the front end of a potentially great virtual organ. This end (product) seems to be based (in some cases) on lack of a physical (good) space and then high quality speakers (along with quality amplifiers) as well as lack of funds. And I know many discussions have taken place on this important part of the equation.

Taking a complete system out of one location and erecting it (exactly) in another quite obviously will have different results just as moving a real P/O does. Again, in several discussions I've had recently, this point seems to have been missed. It appears that some feel, all of these variations can and should be handled in some "magical" way electronically.

Rgds,
Ed

"It appears that some feel, all of these variations can and should be handled in some "magical" way electronically."

I believe that's what Bose does with its upper end systems where you play a CD and the system "listens" and adjusts the frequency curve and output of each speaker to the room it finds itself in. Rather marvellous but you certainly pay for the magic. I imagine it is somewhat easier if the system is a closed one in which all the parts (the Bose speakers and amps etc.) are known beforehand.

[mdyde]

Hello Joe,

P.S. I'll send an email to Dr. Pykett now to see if he would mind giving us a brief response to this topic.

Edit: I've sent him an email.

I believe that's what Bose does with its upper end systems where you play a CD and the system "listens" and adjusts the frequency curve and output of each speaker to the room it finds itself in. Rather marvellous but you certainly pay for the magic

However, such room correction facilities only apply at a point (defined as less than a wavelength, so much less than the distance between the ears at HF). Applied literally, they can also make bass much worse. Spending the same extra money on proper acoustic treatment of the room will always be the better bet, so long as it is possible (and allowed!) in terms of space and decor.

Paul

Regarding room treatment, since "upgrading" my system, I find I have a few "hot spots" which need correction (and a few standing waves as well). Looking into a solution(s), I found some interesting, rather economical approaches that aren't difficult or costly to implement. Even tho we may want a relatively reverberant space, in a smaller room setting, that can pose some issues.

One corrective action is in the form of acoustical panels that can be easily built and moved around. One "sell to wife" point is that these panels do no more "damage" than hanging a picture and with the proper covering, can be considered a decor piece. Check out: http://www.atsacoustics.com/cgi-bin/cp-app.cgi

http://www.atsacoustics.com/page--Free-Online-Room-Acoustics-Analysis--ora.html

http://www.atsacoustics.com/item--ATS-Acoustic-Panel-24x48x2--1025.html

or what I plan to use: http://www.atsacoustics.com/cgi-bin/cp-app.cgi?usr=51H4195712&rnd=9944265&rrc=N&affl=&cip=172.162.99.210&act=&aff=&pg=prod&ref=AB2&cat=106&catstr=HOME: Note, price is for 6 panels. This product has the same specs as does Owen Corning 703, but less expensive.

I will cover them with:http://www.atsacoustics.com/item--Burlap-Fabric-56-inch-wide--1009.html Total shipping is quite reasonable also.

In the "how-to-do-it" department, watch vidoc: http://www.youtube.com/watch?v=-Lc0BjFHoA8

We really need to consider the "last mile", that being the room itself. If the room isn't "right", the other parts of our setup won't be either.

Rgds,
Ed

I love a good debate - even if I am on the losing side as I may well be in this instance. A chance to learn!

Martin, I truly appreciate the amount of effort and thought that went into your analysis. I am admittedly way out of my depth here, but as I understand your argument, superposition (or additivity) means that the integration of infinitely many individual sinusoidal responses equals the original stimulus, assuming a linear transfer function. Does this therefore assume that the transfer functions for air and an electronic signal processor (e.g., an amplifier) are equally linear and behave in exactly the same way?

The aspect of this that is troubling for me (I am honestly not just trying to be pig headed) is that personal empirical experience with mixing (layering) many individual sounds electrically does not ever produce that same result as doing so acoustically. To sample a string section, it is not enough to record twenty violinists one at a time and then layer these individual recordings in a mix. It will sound nothing like having recorded the twenty violins as an ensemble. I believe the key difference lies in the manner in which sounds interact and add in the air compared with an electronic summation. I humbly ask, what is there about the presence or absence of the reverberation in the organ samples that influences signal mixing effects for the sustained portion of the sound? A dry sampled organ may be the very same instrument but recorded with the microphones positioned rather more closely to the pipework and the reverberation release quickly faded out.

I am under the impression that Dr. Pykett's discussion of signal mixing is irrespective of the manner in which the sounds were recorded, i.e., wet vs. dry; only that when similar signals are summed electrically there will be objectionable interference effects. I do hope he will weigh in. I have traded a few emails with him, but he has asked that his comments remain private, so I have respected that.

With kindest regards,
Joe

[mdyde]

Thanks, Joe.

No problem.

I am admittedly way out of my depth here, but as I understand your argument, superposition (or additivity) means that the integration of infinitely many individual sinusoidal responses equals the original stimulus, assuming a linear transfer function.

Correct.

Does this therefore assume that the transfer functions for air and an electronic signal processor (e.g., an amplifier) are equally linear and behave in exactly the same way?

Also correct. Summing the samples digitally (e.g. in a computer) is of course 100-percent linear (computers don't make adding mistakes!). Addition of sound/pressure in air is also nearly perfectly linear (within the range of frequencies and sound levels relevant for organs/music).

The sound card's DACs, the amplifier, and the speakers will all introduce small non-linearities. However, no more or less so than listening to a (wet-recorded) CD recording of real organ (or other piece of classical music recorded live with a single fixed pair of microphones).

So playing back a good wet-sampled organ through a computer and a single pair of headphones or near-field studio monitors should sound no worse than listening to a good 'live' (wet) CD recording of the same organ through the same amplifier and speakers/headphones.

I.e. if good commercial organ CD recordings sound good to you on your amplifier/speakers/headphones then a good wet-sampled organ will sound just as good (because the non-linearities of the amplifiers/speakers will be identical in both cases, so have exactly the same effect on the sound).

[Non-linearities in the microphones, mic pre-amps and ADCs used to record the organ will result in slight differences between a summed wet-sampled organ and a corresponding live recording (because those non-linearities will accumulate with the number of samples playing). Hence for sampling an organ it's essential to use microphones, mic pre-amps and ADCs that are as linear as possible. However, that's not relevant to Dr. Pykett's 'signal mixing effect' because even with playback through a separate speaker (or impulse response) per pipe with 'in air mixing' the effect would be identical.]

The aspect of this that is troubling for me (I am honestly not just trying to be pig headed) is that personal empirical experience with mixing (layering) many individual sounds electrically does not ever produce that same result as doing so acoustically. To sample a string section, it is not enough to record twenty violinists one at a time and then layer these individual recordings in a mix. It will sound nothing like having recorded the twenty violins as an ensemble.

Try doing it wet!:

Set up a stereo pair of microphones (or binaural head) and record those twelve violins separately. Now record the same piece with the musicians remaining in the same places. Keep the microphones in exactly the same positions throughout all of the recordings.

Violins are probably not actually an especially good test case because the individual violinists will almost inevitably adjust the way they play to some degree when playing in ensemble with other violinists, whereas for an organ that's not an issue (although there may be pipe interactions through the wind system and acoustic coupling etc.). So maybe better just to try the experiment with all of the pipes that would sound in a particular chord on an organ.

Now mix all of the sampled recordings together in an audio editor (e.g. put them on separate tracks in an audio sequencer so that you can tweak the alignments easily) and use headphones to compare the result with the live-recorded ensemble version.

As long as there was no interaction between the sound sources (violinists/pipes), and that the sound sources produced exactly the same sound in both cases (and disregarding microphone non-linearities), then you shouldn't hear any difference between the ensemble recording and the digitally-summed sampled version.

(Note also that strictly this assumes that the sounds started at exactly the same instants, relative to each other, in both cases, so you might need to tweak the alignments of the samples. For pipe organs this involves attempting to capture and recreate the individual delays between a key press and the sounding of each of its pipes.)

(Also, in practice, you will never get exactly the same sound from a given pipe on a real organ, so phase relationships etc., and resulting phase addition/cancellation, will vary slightly from one key press to the next - whether a live ensemble recording or with samples.)

I believe the key difference lies in the manner in which sounds interact and add in the air compared with an electronic summation.

Nope - the key factors are the positions of the microphones and of the sound sources. If you keep both fixed (=wet sampling) then the maths and acoustics work properly and (ignoring the slight non-linearities of the microphones) you get the same result from summing the sampled version as you do from recording an ensemble version.

I humbly ask, what is there about the presence or absence of the reverberation in the organ samples that influences signal mixing effects for the sustained portion of the sound?

Likewise, all that really matters is that the microphones and sound sources remain in absolutely fixed positions throughout the whole sampling/recording process (preferably with the microphones in a good listening position!). Then the frequencies and phases will in add in exactly the same way, whether sampled or recorded as an ensemble, giving exactly the same result.

The original room acoustic affects the frequencies and their phases (transfer function / impulse response) recorded from each pipe, but as long as the positions of the mics and sound sources remain fixed then that gives an identical result in both cases.

Dr. Pykett's signal mixing effect is really just a by-product of the microphones being moved for dry sampling, which causes the frequency and phase relationships to change between the pipes, making it no longer mathematically valid to sum them. I.e. if you sum pipe samples made from different microphone positions and mix them together then you no longer get the same result as recording those pipes from a fixed microphone position (whether sampled or recorded as an ensemble) - that's the 'signal mixing effect'.

Using lots of speakers or impulse responses helps to mask/minimise it (by randomising phases etc., simulating the behaviour that a real acoustic would have imparted), but fundamentally the issue is that the microphones are usually moved around when recording dry samples.

I actually originally developed Hauptwerk specifically because of this key point, i.e. because only with wet sampling (or in the longer-term future one impulse response per dry pipe) can you reproduce a real organ accurately in its original acoustic, and with the pipe samples summing together correctly to give an ensemble that sounds the same as a live recording of the real organ. I never felt that any of the existing (dry) digital organs had achieved that to my personal satisfaction. (At the time to my knowledge there weren't any other samplers that were designed around that principle (suitable for organs, at least), so Hauptwerk was developed to support the huge polyphony the approach needs.)

Of course, over the years the Hauptwerk world has grown to encompass lots of other approaches and applications too, which is fabulous, but the fundamental reasons why, for home listening (headphones or near-field monitors in a dry listening room), only wet samples (or dry samples with a separate impulse response per pipe) can reproduce a real acoustic 'perfectly' (almost!), or reproduce an ensemble 'perfectly', remain unchanged.

I am under the impression that Dr. Pykett's discussion of signal mixing is irrespective of the manner in which the sounds were recorded, i.e., wet vs. dry;

Phase addition/cancellation will certainly happen when playing wet sample sets, but so will they with a live CD recording of a real organ, or when listening to a real organ in the flesh. If the wet sample set is recorded properly then the end results should be (almost!) identical in each case.

So yes - you could say it does affect wet sample sets, but it affects real organs heard live too, and identically.

Martin,

Thank you once again. You make a very compelling case. Some convincing evidence in support of your position can be found by listening to the Daniel Cook demos performed on the Father Willis which were recorded directly to a WAV file in stereo. http://www.milandigitalaudio.com/salisbury-demos.htm. Very nice indeed! It would make for an interesting comparison if Cook had recorded the same music on the actual organ.

Still, some questions remain:

1) Must all ranks be played in their original pitch? I duplicated Pykett's signal mixing demonstration using the OpenDiapasonLarge8 060.WAV file from the St.Annes Moseley. With Sound Forge 9.0, I used this mono WAV file to create two tracks. The right track was repitched by one cent. When the two tracks are played using two speakers, there is no audible degradation. After digitally mixing the two tracks into one, the sound is very obviously degraded.

http://sites.google.com/site/vtheatreorgan/Home/files/TwoDiapasonsSeparate.wav
http://sites.google.com/site/vtheatreorgan/Home/files/TwoDiapasonsMixed.wav

2) Must wet sampled organs be free of noise reduction when processing the samples? Noise reduction always alters the waveform and this would seem to render the components no longer identical to their original form and therefore superposition is no longer mathematically valid. If so, this pretty much rules out wet sampling a theatre organ because the higher wind pressures and installation practices cause them to be universally quite noisy. This brief clip of a Clarinet on Jelani Eddington's website is very typical of the noise heard in a theatre organ:

http://www.theatreorganarrangements.com/mp3/Clarinet.mp3

Perhaps somewhere in the world there exists a noise-free TPO in a room with great acoustics, but I have never found such an instrument. A few years ago I did try to wet sample a ten rank Barton in a local theatre. After investing many hours in this project I had to toss the entire lot out.

Best Regards,
Joe

[mdyde]

Hello Joe,

Some convincing evidence in support of your position can be found by listening to the Daniel Cook demos performed on the Father Willis which were recorded directly to a WAV file in stereo. http://www.milandigitalaudio.com/salisbury-demos.htm.

Yes - probably the simplest and quickest proof that wet sampling does actually work from a theoretical point of view is just to listen to the results! If it didn't work properly, then stops wouldn't combine naturally, so the plenums wouldn't sound completely natural (because of the 'signal mixing effect').

E.g. the demos labelled with 'original acoustics' on our demos page were all recorded from wet sample sets using just a single stereo output via Hauptwerk's built-in recorder:

http://www.hauptwerk.com/learn-more/audio/

1) Must all ranks be played in their original pitch? I duplicated Pykett's signal mixing demonstration using the OpenDiapasonLarge8 060.WAV file from the St.Annes Moseley. With Sound Forge 9.0, I used this mono WAV file to create two tracks. The right track was repitched by one cent. When the two tracks are played using two speakers, there is no audible degradation. After digitally mixing the two tracks into one, the sound is very obviously degraded.

(Although I made it myself, I definitely wouldn't recommend using St. Anne's as any kind of test case, since it was the very first sample set ever made for Hauptwerk, essentially as an experiment, it's mono, and was made using relatively inexpensive recording equipment, and has gone through a lot of experimentation and processing over the years. So it's by no means a good model for wet sampling technique!)

However, the experiment you mention there is essentially the same as the one in Dr. Pykett's article.

Frequencies and phases add and cancel in real rooms too (think of room nodes reinforcing bass at particular points in space as an illustration).

That experiment (and yours) really just demonstrates that if two separate sounds are heard (or recorded) simultaneously emanating from different points in space then the results will be *different* to the same two sounds emanating from the same point in space. No dispute there.

What matters is what the *correct* result is, i.e. whether what you hear by mixing the samples is or isn't identical to what your ear would have heard if you had played the real pipes together.

If you take two properly-recorded (not St. Anne's!) wet pipe samples and mix them together (synchronised correctly) then the result should be (almost exactly) the same as having recorded the two pipes sounding together.

If you played back those wet two samples through different speakers or headphones channels then the result would certainly be *different* (as Dr. Pykett's experiment demonstrates), but his experiment makes no comment about which is actually the *correct* one. In this case (for a properly-recorded wet sample set), the digitally mixed version should be the correct one, and playing the wet samples back through two separate speakers would be incorrect from a theoretical point of view (regardless of which might happen to sound nicer for any particular samples).

Consider the following experiment:

You have two sound sources (speakers), each at different physical positions in a room, each of which plays a pure sine wave, one at a frequency just a little above the other, and each one triggered from a different key on a keyboard, so that they can be played like two simple organ pipes.

Now keep your head in a fixed position, keep one ear covered (so that we only have one ear/point in space to consider for simplicity), and play one key, then the other, then play both at once (pressing the two keys at exactly the same instant).

You will hear phase cancellation/addition when they sound together because the sound waves will add (in air) at the point in space at which the two sounds reach your ear.

Now perform exactly the same procedure, but with a microphone kept fixed in exactly the same position at which you had your ear. Also this time, imagine that the keys record a small direct electrical synchronisation pulse when they are pressed (so that you can line them up properly later). (For example, you could use an audio sequencer like Sonar to record the microphone's output, as well as the MIDI messages from the keys, so that you could use the MIDI messages as the synchronisation points.)

If you listen to the recording then you would hear, as expected, exactly what your ear heard the first time (ignoring any slight colouration by the microphone/headphones/etc.).

Now extract the individual note recordings, and trim them so that they start at exactly the moment that the electrical/MIDI synchronisation pulses for the key presses were recorded.

Mix those two recordings (samples) together in an audio editor, without changing them in any other way (no amplitude adjustments, etc.).

Now compare that mixed recording with the recording excerpt of the two sound sources sounding together.

You should find that the results are identical.

2) Must wet sampled organs be free of noise reduction when processing the samples? Noise reduction always alters the waveform and this would seem to render the components no longer identical to their original form and therefore superposition is no longer mathematically valid. If so, this pretty much rules out wet sampling a theatre organ because the higher wind pressures and installation practices cause them to be universally quite noisy.

The key assumptions for wet sampling are as follows:

Provided that:

1. the microphone/ear didn't move throughout the entire recording process, and that:
2. they were synchronised properly, and that:
3. no other processing is performed on the samples which might affect amplitudes or phases

... then the relative phases and amplitudes of all frequencies (and thus pipes) will be identical whether recorded/heard together or separately, so they will sum to give exactly the same result.

I.e. the amplitudes and phases of all frequencies will add and cancel in the same way *under the above three conditions* when heard together, regardless of whether they were mixed in air or in an audio editor.

(Those are the three assumptions by which I define 'wet' sampling, rather than the term making any particular comment about how reverberant the original room is.)

Any processing that you do that affects any audible frequencies or phases will make the digitally-mixed plenum different to a live-recorded plenum to some degree.

However, the frequencies and phases produced by all real organ pipes vary constantly to some degree anyway (even during the steady-state speech of pipes), so small modifications to samples for noise reduction etc., aren't necessarily a cause for concern. Maintaining phases accurately (e.g. using zero-phase filters) in any processing is preferable because it will help to preserve phase relationships when the samples sound together (hopefully good noise reduction software would do that anyway).

In practice, the small changes that result from things like careful noise reduction to remove the noise introduced by a (relatively quiet) blower are probably less relevant to the effectiveness/accuracy of the plenum than other non-linear effects in the real organ, such as 'pitch pulling' (which can't be so easily modelled with sampling). So there's a fair bit of leeway in practice.

Perhaps somewhere in the world there exists a noise-free TPO in a room with great acoustics, but I have never found such an instrument. A few years ago I did try to wet sample a ten rank Barton in a local theatre. After investing many hours in this project I had to toss the entire lot out.

Yes - it could well be the case that practical considerations (like the level of blower noise, or the desirability of the real organ's acoustic environment) render wet sampling not a viable option for a theatre organ.

[mdyde]

Hello again Joe,

Dr. Pykett replied with the following, and kindly gave us permission to post it here, although he intentionally doesn't participate in forums personally (so as not to compromise his impartiality):

I can't see that the signal mixing issue has much to do with whether the samples are wet or dry. If they are wet, one nevertheless still has merely a single signal, statistically stationary in terms of amplitude and phase for each channel (L or R) for each sample emerging from the synth once the reverb has stabilised during the steady state phase of the sound, provided the note remains keyed for long enough. This signal has a defined harmonic structure in which each harmonic has a certain amplitude and a phase relative to some arbitrary datum. Other than minor random fluctuations, which also occur for a real organ pipe sample recorded in dry conditions, or synthesised artificially so as to include them, the amplitudes and phases of each harmonic remain pretty much stable for the purposes of this discussion.

If such a signal is then mixed (added) electronically with another of similar characteristics, then audible interference (in the technical sense of the word) will occur between every pair of harmonics with identical or near-identical frequencies in the two signals, as I demonstrated in my web article. When there is zero relative phase shift between each such pair, the resulting mixed signal for each harmonic pair will have an amplitude maximum, and when the phase shift is pi (180 degrees) it will be minimum. If in addition the amplitudes of each harmonic pair are the same, they will completely cancel in the latter case, as in the audio example on my website. Therefore, none of this has anything to with whether the samples are wet or dry.

For those who remain unconvinced, why don't they just try repeating the experiment described in detail on my site but using samples of any degree of wetness they choose? Or maybe you could do it for them and post the results for all to hear, as I did.

To repeat, note this has nothing to do with wet or dry samples once the signals have reached their steady state in the recording auditorium. There will of course be different, transient, effects during attack and decay due to the attack/release transients of the pipes themselves plus whatever impulse response characterises the particular environment they were recorded in, but a short time after key-down these effects will vanish. It is only during the steady state sound regime that phase interference effects are subjectively relevant for the organ.

Hope this helps.

With the greatest respect for him (and the many excellent and informative articles he provides), I think he has actually missed the key point (which could well have been my fault for not making it clearly enough in my original mail to him), viz.:

That two wet pipe samples when added together give an identical result to a recording of the two pipes sounding together provided that the wet samples were recorded according to the rules for wet sampling i.e.:

1. the microphone/ear didn't move throughout the entire recording process, and that:
2. they were synchronised properly, and that:
3. no other processing is performed on the samples which might affect amplitudes or phases.

However, Dr. Pykett doesn't wish to be involved in further discussion (or to respond to that specifically), so we must respect that and leave the question of how he sees that in relation to his 'signal mixing effect' not completely answered.

Moving on:

Speaking as somebody who has spent the majority of the last ten years of my life developing Hauptwerk based on the one key fact that wet sampling *does* work properly (i.e. that pipes can indeed be combined properly without the 'signal mixing effect' distorting the result when combined), believe me: it is a point I know very, very well indeed to be correct! Likewise it's the single fundamental fact upon which all wet sample sets, and all wet sample set producers (MDA, OrganART Media, Sonus Paradisi, ...) absolutely rely. If it didn't work properly, i.e. if the 'signal mixing effect' did mean that wet pipe pipes couldn't be combined digitally, then none of us would have been doing this, and the audio results you can hear wouldn't have been possible:

http://www.hauptwerk.com/learn-more/audio/

To prove once and for all that wet sampling does indeed work properly, i.e. that it allows multiple separate pipe samples to be combined digitally to give the same result as a recording of the real pipes sounding simultaneously, I've spent today carrying out the experiment I described to you in my previous post. Everything is included in this ZIP, including the source sample files, and the resulting audio files:

http://downloads.hauptwerk.com/forum_temp_images/SignalMixingWithWetSamplingTechnique-AirVsDigital.zip

Open the .cwb file in Sonar to see the test. The first track is the signal that was fed to two loudspeakers (left to one, right to the other), placed at arbitrary points in a room.

The second track is what a microphone (placed at another arbitrary point in the room) picked up whilst the first first track was playing via the loudspeakers. (Only one (mono) microphone was used, so both channels of track 2 are identical.)

The input (track 1) contains, in order:

1. Test1Input-L.wav (a mono middle C diapson sample, fed to speaker 1)
2. Test1Input-R.wav (the same sample, repitched up 1 cent, fed to speaker 2)
3. Test1Input-LR.wav (both of the previous two samples, sounding together, each through their respective speakers, i.e. mixing 'in air')
4. Test2Input-L.wav (a mono middle C diapson sample, fed to speaker 1)
5. Test2Input-R.wav (the same sample, repitched up 5 cents, fed to speaker 2)
6. Test2Input-LR.wav (both of the previous two samples, sounding together, each through their respective speakers, i.e. mixing 'in air')
7. Test3Input-L.wav (a mono 500 Hz sine wave, fed to speaker 1)
8. Test3Input-R.wav (a mono 505 Hz sine wave, fed to speaker 2)
9. Test3Input-LR.wav (both of the previous two sine waves, sounding together, each through their respective speakers, i.e. mixing 'in air')

TestOutput-Complete.wav is what the mic picked up in its entirety whilst the above inputs were played in sequence. (It is the whole of track 2 from the Sonar project exported as a .wav file, and converted to mono, since there was only one microphone anyway.)

I used Sonar to play back the samples and to record the result in this way so that perfect synchronisation could be guaranteed throughout (using Sonar's timestamp), to avoid the need additionally to record MIDI synchronisation pulses.

I then cut up TestOutput-Complete.wav into nine parts, each starting at Sonar's timestamp for the start of the corresponding input signal (thus ensuring that synchronisation, and thus all relative phases, were preserved properly).

This gave the following files (in order corresponding to the input files above):

1. Test1Output-L.wav
2. Test1Output-R.wav
3. Test1Output-MixedInAir.wav
4. Test2Output-L.wav
5. Test2Output-R.wav
6. Test2Output-MixedInAir.wav
7. Test3Output-L.wav
8. Test3Output-R.wav
9. Test3Output-MixedInAir.wav

Using an audio editor, I then mixed the whole of:

- Test1Output-L.wav with Test1Output-R.wav to give: Test1Output-MixedDigitally.wav.
- Test2Output-L.wav with Test2Output-R.wav to give: Test2Output-MixedDigitally.wav.
- Test3Output-L.wav with Test3Output-R.wav to give: Test3Output-MixedDigitally.wav.

If wet sampling works properly, i.e. if it indeed is mathematically valid to sum two individual wet samples together with the 'signal mixing effect' producing a distorted result, then:

- Test1Output-MixedDigitally.wav should be identical to Test1Output-MixedInAir.wav and
- Test2Output-MixedDigitally.wav should be identical to Test2Output-MixedInAir.wav and
- Test3Output-MixedDigitally.wav should be identical to Test3Output-MixedInAir.wav.

Open them in an audio editor and compare them for yourself! They are indeed identical.

If you still doubt that it works, try the whole experiment yourself - you will get the same results.

I rest my case!

Having now spent almost three days of work in total replying to this topic and your previous (indirect) one on the subject of the relevance of 24-bit for sample sets, I'm going to leave this discussion here, I'm afraid (otherwise also Brett will be quite justified in killing me). Of course, feel free to continue the discussion, but I wanted to demonstrate once and for all why wet sampling is indeed valid (which I believe the above test results show absolutely conclusively), and I'll have to leave you to your own devices now.

Thank goodness this is over Martin.
When HW 4 comes out we now know it will be three days later than it could have been.
I for one think you deserve a 3 day vacation after all this but that would make HW 4 six days late so I'm simply going to have to veto any vacation plans Martin. Sorry about that.

This was a very interesting discussion for me, but all along I was thinking the same thing, Martin's time was being taken away from working on HW4.

I've never experienced the signal mixing problem that Collin describes but am still bothered by the beating thirds and maybe fifths. So far it seems that multiple speakers really don't help this.

With all the talk that there was about the Mackie HR824's sounding so good, I've been wanting to get a pair myself to compare against what some call the poor man's HR824's, the Behringer B2031A's.

I was not prepared to spend what they normally sell for as used, but the other day stumbled across a pair on e-bay where one speaker wasn't working, buy it now for $225 with free shipping. I just couldn't pass this up as I'm an electronics technician anyway, and I figured even if a driver is bad, I'm still in the cheap.

I found a very comprehensive repair manual I downloaded for $5 that Mackie wrote, very nice. But I do have concerns buying replacement drivers. It appears the only ones available now are the newer ones used in the Mk-2 series that are made in China.

Oh well, we'll see. Once I get these and get them both working, I'll do some comparisons between them and the Behringer's. I'm really curious, and hope the Mackies do sound better, even though the Behringer's really sound good already. I'm hoping the HR's might help clean up the beating intervals a little I hear so readily, and just in general sound cleaner, with less distortion.

Eric

Just tried Martin's test They sure sound the same to me!

I can fully understand and it makes sense so long as the microphone is in one (the same position) for both tests, that they come out the same. But if the mic was moved each time in front of each pipe as is done with dry sampling / standard digital organ sampling, then I can see there would be a big difference, comparing this to a recording of all pipes playing at once into the microphone sitting in one spot.

Thanks again Martin, just shows how wonderful HW is .