The Next Level of Hauptwerk Realism?

Recently I’ve been reading some interesting resources online, and thinking about next steps for Hauptwerk. Here are some websites that I have been reading, that many of you would be familiar with:

http://lawrencephelps.com/indexa.html

http://www.pykett.org.uk/completed_work.htm

http://cbfisk.com/charles-fisk/

One point that Lawrence Phelps makes in many of his articles is that the sound of a combination of pipes is more than the sum of its parts. He argues that when several pipes are winded from the same wind source, it affects the amount of wind each gets, how the wind goes through the pipe, etc. I would also add that the acoustic mixing, including adding and subtracting of sound waves in the air, is also distinct for multiple pipes sounding at the same time. Now, most Hauptwerk sample sets are made up of each rank or set of ranks sampled chromatically. That means that when more than one stop is drawn, several independent wav files are played back at the same time. This approximates the sound, however I would argue that the true sound, or a more authentic sound is a recording of the same stops sampled when drawn together. That means that more realistic than playing multiple recordings at once is playing a single recording of those same stops. As Colin Pykett also observes, this would also reduce the amount of intermodulation distortion, particularly for those who listen to Hauptwerk in Stereo, which is still the most popular way to listen to Hauptwerk.

The main problem of course, is the amount of extra RAM such a sample set would take up. However, not every combination necessarily need be sampled. Commonly used combinations, such as the stops that make up a principal chorus, or the 8 and 4 foot flutes +/- the reeds in the division, the Plein jeu or Grand jeu, the 8 foot stops together on a romantic organ, would be more than necessary for most uses. The challenge, however, would be to program an ODF that allows to dynamically switch which samples are used based on what stops are drawn. I’m not sure if this is possible with the current version of the ODF, but it definitely would not be possible in the CODM as it now is. The other aspect, is that sample sets would need to be recorded new, or re-recorded, to take advantage of this technology, since none currently does this.

Part 2 of this conjecture is what Charles Fisk calls “articulateness” of an organ. Basically, when more than one pipe is winded from the same source, and a note is held, then another note is played, it causes a waver in the wind supply to the note that is being held. This helps our ears detect the rhythm of the notes, but is not universally present in all organs (though definitely in baroque era organs with slider chests, and modern examples built after that design). To increase the realism in these kind of organs, in addition to sampling each pipe’s speech and cut off, it would also make sense to sample each pipe as a second pipe is played, and after the onset waveform, the recording of both pipes together. This will accurately reproduce the articulateness of some organs. As with the previous example, not all combinations of keys necessarily need to be sampled, but I think it would make sense to sample the most commonly used combinations (say, all the 3rds, or the notes within a 5th of the other note, or all the notes within a 5th in commonly used keys). I’m sure the wind model simulates this to some extent, but again, the greater realism would be to reproduce it in samples. Again, I also don’t know if the ODF would support this, but I think it would result in a greater level of realism if it could.

Perhaps some would ask why this is necessary, and I would admit it is not strictly “necessary” to the enjoyment of Hauptwerk. However, I think it would introduce a greater level of realism, and for a small sample set, the extra RAM consumption wouldn’t be too onerous. Perhaps others don’t agree with the opinions of Phelps, Pykett, and Fisk on these subjects, but I think it makes for an interesting discussion about the future of Hauptwerk. Perhaps sample set makers will balk at the extra work that making such a sample set entails, but if one does it and it becomes a popular set, would it be worth the effort then? Perhaps these subjects have been discussed before, but in the light of frequent Hauptwerk updates and greater availability of large amounts of RAM, I think they are interesting to discuss now. What do others think of these points?

Can't argue against a recording of many combinations would sound realistic in stereo, though I'd find pipes moving around between stereo pairs because of joining different combinations a bit odd in multichannel.

While the goals are great, I'd personally not buy more than my current 64GB of RAM to use a larger sampleset with extra samples for common combinations. It takes too long to load from SSD on a good 8 core machine for me to put up with anything over 50 - 60 GB. Also I'm finding prices on large samplesets harder to justify to myself these days.

I do think an early sampleset from Pipeloops (?) had sampled combinations such as fonds 8, 8 4, 16 8 4 on it, but I don't remember which one.

Windchest-level implementations modulating per-pipe samples are probably reasonable without consuming vast compute resources or raising prices too much.

Some of that's already done in HW. For example, holding a bass note and then adding a high note perturbs the smaller pipes noticeably on some baroque samplesets like Freiberg where wind effects are really audible.

SP recently introduced a pipe coupling feature with his latest Doesburg revision, which is another good step in wind modelling. Maybe something along those lines to simulate pitch draw between nearby pipes could eventually be done in HW.

I'm happy how my samples sound now and am very pleased to be able to play organs from around the world at home.
All this sounds interesting as long as new sample sets are kept within sensible RAM limits.

[mdyde]

Hello larason2,

I'm afraid I don't really have time to be involved in this discussion, but very briefly: It would be technically possible currently for a sample set producer to make a sample set that used/substituted sampled combinations (or even samples or every possible key/stop combination), by using appropriate switching in the organ definition, if he/she so wished. There are of course some major disadvantages in doing so:

- You would lose the ability to adjust the tuning or voicing of individual pipes/ranks.
- You would lose the ability to apply real-time per-pipe modulations (e.g. wind supply model fluctuations, speech fluctuations, tremulants, and swell boxes) to pipes individually.
- Recorded combinations can sound rather static, since their individual pipes' pitches and phase relationships are fixed within the samples, and cannot change over time (e.g. for subsequent key presses, unless you recorded multiple samples for each possible sampled permutation, picking them randomly upon playback, which Hauptwerk can do if desired). Hauptwerk constantly modulates the pitches, amplitudes, and brightness of all pipes separately in real-time to avoid that.
- You would lose the ability to route pipes separately amongst speakers (and/or for applying impulse response reverbs).
- Sample sets would be vastly larger, and take correspondingly more time/money to make.

Trying to record into samples (rather than model, via Hauptwerk's existing wind supply model) wind supply fluctuations would, in my view, by impractical, since, for example, a bass pedal could be played by the organist part way through the sounding of a note held on a manual, or at any other point in time (during, or prior to, any other note). Wind pressures fluctuate constantly throughout the organ, depending on what precise combinations of pipes have been played, and at what points in time (e.g. how recently) and for what durations. Hauptwerk's existing wind supply model handles all of that (using fluid dynamics modelling).

SP recently introduced a pipe coupling feature with his latest Doesburg revision, which is another good step in wind modelling. Maybe something along those lines to simulate pitch draw between nearby pipes could eventually be done in HW.

Yes -- we do have an enhancement request logged (actually since v2) for including a native model for pipe acoustic coupling.

There are two early sets from Pipeloops that use recorded combinations of stops:
Madeleine (Paris) and Laeken (Belgium).

The Madeleine sampleset contains just 3 combination for the manuals and 2 for the pedals.
For the manuals: Fonds 8', Fonds 16'/8'/4' and Tutti.
where Fonds 8' stands for the Fonds 8' for all the manuals coupled, etc.

With a cachefile of just half a GB it loads in seconds and the sound is really impressive, even today.

The Laeken set contains not only a lot of sampled combinations but also a number of stops sampled alone
Some of the combinations can also be made by combining the right single stops.
So it is possible to compare these combinations in the way you hear them, as a sampled combination or
as a combination of the apprpriate stops samled alone.

The difference in sound is remarkable.
But the difficulties/ impossibilities Martin just mentioned are surely more important than this sampled "mixing in air".

Regards,
Jack

That's interesting, thanks all for weighing in, particularly Martin. While the technical aspects sure appear to be difficult, I don't think they are impossible. Maybe for Hauptwerk VIII in 2022? The first step will be to implement the same tuning and pipe adjustment to more than one pipe at a time, which I'm sure is in the works. From what Martin says, I'm sure sample set makers aren't going to jump on this any time soon. That's interesting that a sample set already has done this, and the results are very good. I'll have to look into that. I have Georgenkirche by MDA, and a recording of Marie-Claire Alain playing Bach's trio sonatas on the same organ, and while I like them both very much, I also feel that the ensemble sound she gets from the pipes is better than the ensemble sound of the same stops registered on Hauptwerk. I also feel the mixtures on the sample set are a bit brighter and more aggressive. Looking forward to Hautpwerk the next generation!

The discussion is very interesting. But I agree with Martin, that the effort is high and you lose a lot of editing options.I have Madeleine, but for me the difference to other samples is insignificant.For me the ideal way would be reached, if there was no more difference to be heard between the HW sample and a good recording of the real pipe organThis can't be without cost, of course, but it would be worth it to me.

Interesting indeed. I listened to Jiri's audio examples on the SP website of the 3 different excerpts of BWV 532 which are, no pipe coupling, defualt pipe coupling, and maximum pipe coupling. He states that example 3 (maximum pipe coupling) is too far out of tune but is still playable. I'd say it's still really not all that bad, and it does produce quite a chorus effect which in his example is perhaps a bit too far out there and it's time to call in the organ tuner. I'd say a setting somewhere between the default and maximum settings would likely be ideal as I don't hear much of a difference (or enough of a difference) comparing the default setting to the no coupling setting. I've used the random tuning feature in HW with success and it produces much of the same effect depending on how far you take things, if so desired you can make the given sample set sound pretty ridiculous using the HW feature as well. Although Jiri refers to it as being 'static' it does make me wonder between his new pipe coupling feature and what HW does if there would be much of a discernible difference in sound to make it worthwhile pursuing further? I guess I do question the 'static' part as my understanding of the HW tuning feature is it's 'random' and if it is then there's likely not going to be a discernible difference. It's kind of one of those 6 to one, a half dozen to another deals.

Marc

I guess I do question the 'static' part as my understanding of the HW tuning feature is it's 'random' and if it is then there's likely not going to be a discernible difference. It's kind of one of those 6 to one, a half dozen to another deals.

My understanding, which may be incorrect), is that the HW random tuning feature affects an instrument when it is loaded (perhaps differently each time), but then becomes static while you're playing that particular loading instance. There has also been considerable discussion of the fact that, currently, the HW feature cannot associate the different sound perspectives — e.g, near, diffuse, distant, etc.) of a particular pipe so that they are all randomized the same, and in tune and phase coherent with each other. Thus you can potentially get a chorus effect between a single pipe and itself if you tweak it too far with a surround sample set.

My understanding, which may be incorrect), is that the HW random tuning feature affects an instrument when it is loaded (perhaps differently each time), but then becomes static while you're playing that particular loading instance. There has also been considerable discussion of the fact that, currently, the HW feature cannot associate the different sound perspectives — e.g, near, diffuse, distant, etc.) of a particular pipe so that they are all randomized the same, and in tune and phase coherent with each other. Thus you can potentially get a chorus effect between a single pipe and itself if you tweak it too far with a surround sample set.

Good points, Grant. I too am not sure and maybe someone can clarify, but my assumption is / was the HW feature will randomly de-tune here here and there as you go along and not in the same pattern or be static, but I could be wrong. I'm no expert in recording, but in a 6 channel set as an example, to me a simple way of dealing with the HW feature not being able to associate the difference between front, diffuse and rear channels would be to do all of the recordings at the same time and then have the option of blending them which seems to be the case with some of the sample sets such as those from SP. Then if pipe 'A' of the close mix goes slightly out of tune I'd assume then the diffuse and rear would follow and do the same. Now, if the 6 channels are all separate recordings of the same, then I'd agree that the close pipe 'A' would not correspond with the diffuse and rear of the same pipe. Someone with more expertise of exactly how this all works feel free to explain.

As far as the HW tuning feature goes I recently experimented with the Armley Schulze. The default setting is 100 and it does sound almost too perfectly tuned, raising it to 150 is too much, 125 seems about right and does make a nice overall difference.

Marc

One point that Lawrence Phelps makes in many of his articles is that the sound of a combination of pipes is more than the sum of its parts. He argues that when several pipes are winded from the same wind source, it affects the amount of wind each gets, how the wind goes through the pipe, etc.

Agree! So I would say that the sound of a combination of pipes is a product of the pipes and the wind behaviour (and the room acoustics).

The Laeken set contains not only a lot of sampled combinations but also a number of stops sampled alone
Some of the combinations can also be made by combining the right single stops.
So it is possible to compare these combinations in the way you hear them, as a sampled combination or
as a combination of the apprpriate stops samled alone.

The difference in sound is remarkable.

So maybe the conclusion should be that the wind/detuning model in this (old) sample set was not quite perfect.

[mdyde]

I too am not sure and maybe someone can clarify, but my assumption is / was the HW feature will randomly de-tune here here and there as you go along and not in the same pattern or be static,

Hello Marc,

There are effectively two relevant features/adjustments in Hauptwerk:

- Pipe tuning randomisation, which is applied to each pipe separately whenever the organ loads, and then remains static (until it's next loaded).
- Pipe air flow randomisation, which causes the air flow (and thus pitch, amplitude, and harmonic content) of each pipe separately to vary slightly, changing constantly in real-time, thus imparting movement and avoiding static phase relationships when pipes speak together.

(Adding native support within Hauptwerk to allow surround/3D sample channels to be handled together in terms of pitch/phase/etc., to avoid unrealistic chorusing/phasing between surround channels, is a logged as a very high-priority enhancement.)

There are effectively two relevant features/adjustments in Hauptwerk:

- Pipe tuning randomisation, which is applied to each pipe separately whenever the organ loads, and then remains static (until it's next loaded).
- Pipe air flow randomisation, which causes the air flow (and thus pitch, amplitude, and harmonic content) of each pipe separately to vary slightly, changing constantly in real-time, thus imparting movement and avoiding static phase relationships when pipes speak together.

(Adding native support within Hauptwerk to allow surround/3D sample channels to be handled together in terms of pitch/phase/etc., to avoid unrealistic chorusing/phasing between surround channels, is a logged as a very high-priority enhancement.)

Hello, Martin and thanks for the clarification and explanation. Would the way a multi-channel sample set is recorded have any effect or benefit as to how the handling of pitch / phase / etc. would be between the channels?

Marc

[mdyde]

Hello Marc,

In Hauptwerk currently, not really, since sample sets have separate virtual pipes/ranks for the surround perspectives, and separate pipes/ranks can't currently play phase-locked together (but as I mentioned, adding support for that is logged as a very high priority enhancement).

However, I understand that Pipeloops (and perhaps some other producers) include things within their organ definitions to try to mitigate that to try to ensure that the various virtual ranks for a given real play at the same pitches (e.g. by having their own ODF-implemented alternatives for Hauptwerk's pipe tuning randomisation).

Another sample set that includes choruses is the Mietke harpsichord. It’s admittedly a much less complicated affair and sound, but I do believe there to be an audible difference when drawing the two 8’s and 4 separately vs the prerecorded combination. Part of that is the characteristics of the speech of the second manual when they are physically coupled in real life. Interesting nonetheless.