It is currently Fri Mar 29, 2024 3:38 am


Version 2 Tremulant Handling

A discussion forum for anything even marginally Hauptwerk-related.
  • Author
  • Message
Offline

dna

Member

  • Posts: 262
  • Joined: Sun Aug 10, 2003 3:49 pm
  • Location: Fremont, CA

Version 2 Tremulant Handling

PostTue Apr 13, 2004 2:00 pm

Martin,

Back in August of last year I posted a list of questions in the “Help” threads. One of those questions was “Are there any problems using tremulant samples (for example, cross fading to release)?” You responded:
Of course you can get Hauptwerk to load them, but I think it's far from an ideal solution. The main problems are:

- the phase of the tremulant won't align when cross-fading to the release, as you point out
- the phase of the tremulant of each pipe that is playing simultaneously won't match
- you cannot model the spin up/down of the tremulant.

Version 2 will include wind supply modeling, which will affect the pitch and amplitude, and also allow for wind pressure-determined cross-fading between layered samples (this simulating periodic harmonic changes). I'm hoping that this should give a very realistic tremulant model, without the shortcomings of using tremulant-effected samples.
In another thread we have been discussing the best way to implement theatre organ tremulants. Although I believe I understand the general concept, I’m not clear on how one might implement your wind supply modeling for tremulants.

I assume I start by sampling each pipe with the tremulant on, correct? Then I’d extract short samples from various pressure phases of a tremulant cycle, for each pipe, right? HW2 will cycle through these layers, cross-fading from one to the next, when the tremulant is on. Through trial and error one will determine the number of layers necessary to give a good simulation.

Am I right so far or completely screwed-up? Assuming I’ve got the right idea, how does one handle attack and release samples? Does HW2 cross over at the end of the attack to the current tremulant cycle no matter how discontinuous that might be? Or does HW2 stay on the full-pressure (non-tremulant) sample until the next tremulant cycle starts? Or does one record several attack and release samples as well? The more I think about this, the less sure I am that I’m on the right track.
-David-NA
Offline
User avatar

mdyde

Moderator

  • Posts: 15444
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

PostWed Apr 14, 2004 12:08 pm

Hello David and others who have participated in this thread,

Apologies to you all for my tardy reply - I've been away for a few days over Easter, and I'm just catching up with mails etc.

First of all, the tremulants are probably the most important single aspect of modelling a theatre organ well, so I'll be working with Brett to make sure that, whatever proves to be necessary in the end, the results are as good as we can make them. That said, there is almost always some degree of trade-off required in order to keep processor and memory prerequisites within the bounds of reason, along with the magnitude of the task of sample set preparation. There is no point in a fantastic tremulant if each sample takes a week to prepare, and the computer can only play one pipe at a time!

I stand by my opinion that use of tremulant-affected samples is not a satisfactory option for the reasons stated before. For those reasons, I think that option must be discarded.

Whilst Hauptwerk version 2 can adjust the levels of sample layers according to its wind pressure model, I had not envisaged using lots of 'intermediate' layers as you had described on the preceding thread. A layer can be turned up as the wind pressure increases, with the shape of the amplitude response curve being adjustable, whilst another is turned down, but you cannot presently have an intermediate layer faded in at a certain pressure and then out at a slightly higher pressure. Thus you cannot 'chain' layers in the way you describe.

The other problem is that the phase of the samples within the layers themselves (as opposed to the tremulant phase) would have to be aligned absolutely perfectly, otherwise phase cancellation would occur, resulting in a mess. Because phase must remain locked, the pitch of the layers must remain perfectly matched for all layers comprising the pipe (there is an option for this in version 2), and thus real samples of each part of the tremulant phase would be almost impossible to use: you would realistically have to re-synthesize or use phase-locking algorithms to shape the samples so that pitch and phase were matched. Also, you would need attack and release samples at each possible tremulant phase angle, and also at each wind pressure that would be possible when the tremulant was disengaged (for non-tremulant wind modelling and so that the tremulant could start and stop). Given that each such layer would have to be a full, sustaining sample in its own right, there would appear to be only two ways to achieve it:

1. Record each pipe with a steady wind supply at each relevant pressure, then phase-lock the resulting samples and use them for the layers (which would require disassembly of the organ).
2. Model the effects of each pressure by applying zero-phase filters to the 'full pressure' sample to give a set of filtered samples.

Either way would take many months of work, and require fairly impractical computer hardware for the user since all of the samples would need to be kept in memory and each layer would be consuming a voice of polyphony.

The other problem with approach 1 is that transients present in a tremulant sound due to *changing* pressure would not be recorded. I'm not presently sure quite how much they contribute the sound, but I want to try some further experiments to see.

Approach 2 might be just about usable if there were only two layers - simulated high pressure and low pressure. That was what I had in mind when I originally mentioned layer cross-fading.

However, I think that the most practical, and hopefully adequately authentic, approach is to use a simple real-time filter as part of the wind model in addition to pitch and amplitude fluctuation. Although the version 2 wind/tremulant model is tentatively finished, I will add such a filter if it proves to be necessary to create a good theatre organ tremulant.

The current wind model is really quite comprehensive and probably already addresses many of your points about the tremulant:

- There can be any (realistic) number of independent tremulants, reservoirs, blowers, wind-chests and other objects.
- The wind reservoirs, wind-chests, tremulants and pipes can be connected together in any configuration.
- Flow rates are fully adjustable at every connection point in the model.
- Fully adjustable physical models are used for all moving parts in the model, such as the reservoir table and the tremulant palettes.
- The pitch and amplitude response to pressure is adjustable independently for every pipe.
- Reverb processing (when finished) will happen after wind and enclosure processing.

If I include a filter as well, then its parameters will also be adjustable for each pipe.

My own experiments, together with results I have been able to obtain from various books and papers, suggest that this should give a sufficiently detailed model. Certainly the 'classical' tremulants I have modelled so far seem to be more than adequate, but I think that the addition of a filter may be necessary for a really good theatre tremulant.

From my own experiments, flue pipes increase in pitch with wind pressure by up to about a semitone from the lowest sounding pressure to the point at which they break into over-blowing. Although the overall amplitude increases with pressure, it is mainly the upper harmonics that are accentuated, so I think a low-pass filter should give a pretty good model. Reed pipes seem actually to fall very slightly in pitch as pressure increase, but again the upper harmonics are subject to the greatest increase in amplitude.

The addition of another real-time per-pipe filter for the wind model will reduce polyphony a little further, but I think it can be a relatively simple low-pass filter, so hopefully the impact will be reasonably small and it will be justified. However, more complex shaping of harmonics would be impractical, given that I'm pretty much at the limit of the latest CPUs.

I'll be working with Brett over the next few weeks to see how well the tremulant model works with theatre samples, so we'll have to see what proves best. Maybe we'll post some samples when we have something that we're fairly happy with, so that people can comment.

Hope that makes some sense.

Best regards,
Martin.
Offline

dna

Member

  • Posts: 262
  • Joined: Sun Aug 10, 2003 3:49 pm
  • Location: Fremont, CA

PostThu Apr 15, 2004 11:55 am

Thanks, Martin. Clearly you've given the tremulant issue a lot of thought. Your wind model is very impressive, far more sophisticated than I had imagined.

I’m not sure I understand the advantage(s) of modeling the tremulant, however. Al Sefl writes: “Properly done a trem should be tight and near a sine wave for a pleasing effect. Normally this means 25 cents sharp maximum and no more than 50 cents flat on the deepest Tibia tremulation.” http://atos.stirlingprop.com/kbase/AlSe ... eights.htm If the goal is a sine wave, what do you gain by modeling it?

Does v2's wind model take into consideration pressure wave propagation to each pipe due to distance from its tremulant? That is, can there be tremulant phase differences between pipes? (I don't know if this is a significant issue or not.)

If one wanted to use tremulant samples, in spite of the limitations and issues in doing that, in v2 will it be possible to use a tremulant switch to cause selected stops to use an alternate sample set?

Of course I very much hope your tremulant modeling works well especially since the only other practical solution, tremulant sampling, has so many disadvantages. I’d be more optimistic if I didn’t know how miserably amplitude and pitch tremulants in past electronic organs were in recreating the theatre organ sound, even in very expensive “oscillator per pipe” designs. Maybe what was missing was per pipe tremulant filtering. Do you have any thoughts on this?
-David-NA
Offline
User avatar

mdyde

Moderator

  • Posts: 15444
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

PostThu Apr 15, 2004 12:48 pm

The principal function of the wind model is to model other features of the wind supply, such as wind sag, wind unsteadiness and oscillation of the reservoir table as pipes of different flow rates sound and are released. The tremulant is modelled within the standard wind model mostly because that's how the real thing works, and the mechanism is already there within Hauptwerk 2, so it is convenient and a little more accurate.

You could model propagation of the wind through the wind-chest with Hauptwerk 2's wind model if you wanted to, but it would add some processing overhead (the more wind-containing compartments, the higher the processor load), and I believe that the pressure wave travels at the speed of sound anyway, so I suspect its effect would be completely negligible, and can safely be ignored.

The filters, and perhaps studying and modelling transients present in the tremulant cycle, would have a much more significant effect, I think.

You could easily use tremulant samples if you wanted to. Of course, the big problems remain - the pipes must be retriggered when you engage or disengage the tremulant, and the release samples are likely to sound fairly awful (being at the wrong pitch and sound phase, causing phase cancellation etc.).

Anyway, we'll see how well it turns out shortly.

All the best,
Martin.
Offline

dna

Member

  • Posts: 262
  • Joined: Sun Aug 10, 2003 3:49 pm
  • Location: Fremont, CA

PostThu Apr 15, 2004 1:29 pm

Thanks, Martin. I hope Brett recorded at least a few pipes from each rank with the tremulant on which you can use for comparison and tweaking. If not, I'd think it would be worth the effort to go back into the chamber and do that.

As you can probably tell, I'm excited by what you are Brett are doing and I'm sure many other TO enthusiast will be too once they hear about it. I appreciate the detailed responses too. You mentioned earlier that you might be able to post some tremulant examples later on; I'd be very interested in hearing them.
-David-NA
Offline

dna

Member

  • Posts: 262
  • Joined: Sun Aug 10, 2003 3:49 pm
  • Location: Fremont, CA

PostWed Apr 28, 2004 1:23 pm

I suppose this is obvious to everyone else but it suddenly dawned on me that there is a interrelationship between tremulant and reverb. Anyone who has been inside a theatre organ pipe chamber when the organ is playing with the tremulant on knows that the resulting sound isn't very musical, seems grossly exaggerated and lacks the lush sound heard out in the theatre. (Especially "amusing" is the sound of the tremulated Vox Humana heard up close.) The "in the chamber" effect is similar to earlier electronic organ tremulants that didn't use rotating speakers.

Out in the theatre, sound reaching the listener is mostly reverberant, not direct, and consists simultaneously of sound bounced off various surfaces. What is significant about tremulant sounds, though, is that the phase (pitch) and amplitude is changing and this means that at any instant, the ear is actually hearing a wide spectrum of pitches even from a single pipe as the sound of this one pipe reaches the ear at different tremulant phases at once. And since amplitude of the source is also changing, the direction of the strongest reflections is constantly changing. The end result is that the sound from even a single pipe as heard out in the theatre becomes far more complex and "rich" when the trem is turned on. (This theory also explains rotating speaker's advantage in electronic theatre organs. The usual Doppler-shift explanation is only one aspect of its effect.)

What puzzles me, though, is why artificial reverb doesn't create a similar effect. What I've heard misses the mark by a large margin. Sure, there's the multiple point-source issue that Martin has explained, but I would think that for a single pipe, at least, even a fairly simple artificial reverb should capture the effect I've described above. I can only speculate that 3D realism of reverb is an important part of the theatre organ sound, even more than for for classical organs.
-David-NA

Return to General discussion

Who is online

Users browsing this forum: No registered users and 5 guests