Resonance modelling

I was playing the Immanuel Presbyterian EM Skinner today, and I think I know what it is I'm missing on Skinner sample sets. You see, I've heard a few of those big American-type Romantic organs, both in recordings and in person, and there always appears to be something missing compared to when I play the Skinner. I've heard other people say this too, that nothing (yet!) sounds perfectly like an EM Skinner. The pipes do sound as they should though, it just feels as though there's something missing.

Then I got to thinking about digital pianos. I had an old digital piano (with sample based technology), and the expression pedal on it lasted a comically short time, and it never sounded "right" compared to a real piano. Then, we started getting digital pianos with "modelled resonance," and suddenly the sustain lasted just as long as you would expect, and realism made a big jump. In the case of a real piano, when you play a note, how loud the note is, what harmonics are heard, and how intense the harmonics are is dependent on what other notes are already playing, and how loud they are. The modelled resonance recreates this complex relationship, and it is the technology behind most of the piano VST's and sample based digital pianos that are currently used.

So I think this is what is missing from Hauptwerk, and it's more evident on dry organs, and organs in smaller acoustics. Having convolution reverb is nice, and in part helps to correct for the missing resonance (especially in big acoustics), however, it's not quite the same as hearing the same resonance you would expect in a "real" organ, particularly in dry acoustics. On real American Classic organs, and other organs in a dry acoustic, I feel like in real life they were a lot "warmer" and the sound had more "bloom," and the sound lasted "longer" in a way that was independent of the reverb, which I suspect is the extra resonance. I also think that so far, Hauptwerk isn't really set up to model this kind of acoustic phenomena. It would be great if it was though!

Now, has this been logged as an enhancement request? What does everyone else think? I don't recall this ever being discussed in years past, but maybe it was before my time or I missed it? It would be interesting if it was!

Edit: There was this post in 2019: viewtopic.php?f=1&t=17765&p=132931&hilit=resonance#p132931

But it didn't get very far! I also don't remember it then, I don't know how I missed it!

I wonder if you could explain how the resonance you refer to arises in a pipe organ. I don't doubt it, but what are the causes to be modeled?

In a grand piano, some sources of resonance are fairly obvious: the sound board, the case and lid, any currently undamped strings including high treble with no dampers, the duplicate strings, slight string movements from lifting the dampers, etc. Some of that doesn't get recorded in note-by-note samples, so there's resonance modelling to do to make it sound more like a piano.

An organ might have case or chamber resonance, but that should be in the note samples. Other than wind draw effects, do pipes concurrently playing near each other vs silent have any resonance-producing interactions like open strings do in a piano?

From that previous topic, it sounds as if silent pipes nearby do respond to a speaking pipe, but that should be baked into the recorded samples.

Thanks.

Yes, I think it does. For sure, some of the resonance of nearby pipes is sampled when you play a note, however as mentioned previously, I don't think this is enough, particularly in very dry acoustics. The argument in the previous post was that stimulated resonance in a pipe isn't as loud as the sound caused by wind passing through it, however, I think it still has an effect that is audible, and this effect can be more audible the more notes that are played together, and over time. If you think about a piano, a sampled note also includes the resonance of other strings, so you would think that is enough. However, what resonates and the degree of resonance changes if more than one note is played. In pianos this is modelled, even if the piano sound is based on samples, because the sample isn't enough to produce sounds as we hear them. Also, I suspect that the organ case, the sound board, and the pipes are all part of the resonance, though as mentioned, in a dry environment it probably makes more of a difference.

I had a Casio Privia Pre-resonance modelling that I played for years, but then I played a new one in a store, and I noticed a distinct difference. Here's a write up on one of the new ones with "acoustic resonance modelling":

https://www.soundonsound.com/reviews/casio-privia-px3

Here's a writeup by Yamaha (which is another one of the digital pianos I've heard and played with this):

https://faq.yamaha.com/usa/s/article/000001306

Here is a piano VST that uses samples, but has modelled resonance:

https://synthogy.com/index.php/products ... and-pianos

Again, for all of these, they recorded samples with resonance built-into the samples, but then use a modelled layer on top of that to simulate the extra resonance that arises when more than one note is played together.

Yes, I think it does. For sure, some of the resonance of nearby pipes is sampled when you play a note, however as mentioned previously, I don't think this is enough, particularly in very dry acoustics. The argument in the previous post was that stimulated resonance in a pipe isn't as loud as the sound caused by wind passing through it, however, I think it still has an effect that is audible, and this effect can be more audible the more notes that are played together, and over time. If you think about a piano, a sampled note also includes the resonance of other strings, so you would think that is enough. However, what resonates and the degree of resonance changes if more than one note is played. In pianos this is modelled, even if the piano sound is based on samples, because the sample isn't enough to produce sounds as we hear them. Also, I suspect that the organ case, the sound board, and the pipes are all part of the resonance, though as mentioned, in a dry environment it probably makes more of a difference.

I'm guessing some if not all pipe resonance in a single note sample might have been "denoised" away in some samplesets, making the results more sterile. You certainly make some good points.

I don't have the Skinner, but I have often noticed that the quality of denoising is highly variable among sample sets. The quality of the wind model is another important factor.

Whether a sample set sounds lively and realistic, or lifeless and dull, is often determined by these 2 factors.

This is an interesting topic. I would point out that for pianos the individual strings are very tightly coupled via the soundboard and in particular the bridge. Still one can lift the dampers and merely speak to the piano and generate an amazing resonance. This is especially true with the more resonant (for better or worse) American grands.

It would be a neat experiment to playback sound impulses and tones in or near a pipe chamber and record the resonance from the pipes at the tails of these triggering sounds. If the proposed effect is present it should be clearly audible.

Thanks John. I've actually done that experiment (speaking loudly into a pipe chamber), and can confirm you can hear your voice resonating in the pipes.

I guess where I'm puzzled is, if resonance from other pipes in the speaking pipe's vicinity hasn't been denoised away, is there proof that the effect isn't modeled correctly by linear superposition when you add samples? You can't model an effect without a mathematical model, and I don't know what additional effect we're hoping for, beyond superposition of the resonance in each sample. Is some of the resonance nonlinear and/or time delayed beyond what is captured in release samples?

Yes, I think the resonance is complex, and that is why you need a model. It appears to have both linear and non linear components, involving both the fundamental, as well as harmonics, with not all harmonics being resonated equally. These effects also can be time delayed beyond what is captured in release samples (which is why a sustain pedal note sounds a lot longer with resonance modelling than it does without). For instance, when you yell into a piano, you can hear that some strings resonate primarily the harmonics of your yell, with even low strings strangely resonating mostly high pitches. If you have an acoustic piano, try taking the lid off and yelling at the strings at different pitches. Sometimes you hear low or medium pitches coming from the low strings after a medium pitch yell, some higher pitched yells cause a high pitched ringing from the high strings. Some yells don't seem to have as much coming from the instrument at all.

The way to model it would be to record three notes playing separately, then all three combinations of two notes, then all three together, then compare the amplitude vs. frequency graphs for all of the combinations (on a real organ). The model would have to provide what is missing between the three playing separately, and what we see when all three are played together. This would also have to be done with high and low notes (and combinations of different high vs. medium vs. low). After looking at these graphs for a while, you can see what the primary effects are. Hauptwerk already has different filters for harmonics vs. the fundamental, so once the general trend of an organ's resonance is figured out, it can simply increase or decrease the fundamentals/harmonics (though the harmonics may also have to be divided again based on what is observed).

It may even be possible to program this into an ODF without needing to reprogram Hauptwerk itself (to some extent, not perfectly if certain parts of a harmonic are resonated more than others). It would be a lot of work though!

This may be a bit off topic. I don't know if this is related to resonance modeling, but I will say that the LiquidSonics Illusion product does modulated convolution which to my ears provides extra realism when used in small doses.

I suspect, like others, that as little denoising as possible is another key to realism. I don't see an easy solution but I sure like to see people's original ideas.

I'm sure there is something to the "resonance" thing....

Since I'm the one who de-noised this sample set, I will say that compared to my earlier efforts, I feel it is much less sterile - my first couple of sets I probably did too much damage to the samples during noise reduction, and therefore ended up having a much more "lifeless" sound.

It's a complex art, I can assure you.

Over the last year, I spent quite a few hours working on the Blackburn Cathedral set - it was the most reverberant set I'd ever tackled, and I had to learn some new skills....

Here's the interesting part - Some of the noise reduction settings I had to change in order to preserve the reverb also seem to preserve more resonance in dry sets!

So, on our latest sample set (the Steiner-Reck at St. Luke's McLean, just released) I used that information while doing noise reduction on that set, and I think the difference is obvious.... It's maybe a bit wetter than IPLA, but it's still not a really large reverb.

Here's another interesting thing - when adding reverb via impulse response, you add back in some resonance - a properly recorded impulse response can impart not just reverb, but also resonance. There are quite a few examples of other instruments being sampled (violin, for example) and then the sampler uses an impulse derived from a violin body to convolve the sound and give richness/resonance. It's not reverb... it's the character of the sound being modified by the body of the violin!

I wonder if it would be possible to do something similar within the pipe chamber in order to capture the resonance in a similar way...

Every time I think I have this process figured out, I learn something new..... it is so rewarding to feel like it is possible continually get better.

This thread reminds me of discussions that people had about the importance of an organ case during the organ reform movement. Many churches at that time, especially in the US, had organs with exposed pipes and no case. Proponents argued that the case was important to focus the sound. While there were some builders that continued building case-less organs, most artisan builders encased their instruments.

I'm still wondering why we think HW needs an enhancement to support pipe/case/chamber resonance, unless it's been removed from the samples. I can see the need when samples don't retain whatever resonance was recorded, and there's a desire to model the effect when the real thing was discarded. But otherwise HW does what is needed, as far as I can tell.

Since resonance is audible, it's already recorded in the raw per pipe samples. (Let's assume the release isn't shortened to cut off resonance too early and any denoising only kills background noises.)

Nothing mentioned so far indicates why adding those per pipe signals including resonance shouldn't reproduce the resonance that the listener hears, just like adding wind noises and reverb does.

Is there a study showing that multiple sounding pipes produce a non-negligible, nonlinear resonance response, or was that meant to be a what-if?

At the very least, HW as is would reproduce a good first order approximation from multiple samples, and I'm not sure second order effects requiring a new HW model would be very audible in real life.

The piano sampling situation isn't so applicable for pipe resonance. What breaks down is that, to capture a piano's resonance fully in samples, without modeling, you'd have to record each note and hammer velocity for each position of every damper (both continuous), and, during playback, switch samples on held notes to account for the reduced excitation from decay combined with all the damper movements as they occur. It isn't enough to account for hammer strikes and damper pedal movements alone because each key moves its damper and changes the resonance configuration. There just isn't enough computer in a digital piano to do all that, so it's modeled and mostly not that accurately.

But pipes don't decay while sounding, and organs are pretty much a fixed resonance configuration, not responding differently each time the same pipe is played. Recorded samples and linear superposition should do the job. So why aren't we hearing it? Maybe denoising and release truncation on dry/close samples? Yes, getting rid of ambient noise like HVAC, traffic, blower, and humans is really hard, so maybe keeping resonance in the samples isn't even feasible. And maybe resonance is drowned out by reverb on wet samples?

Using an impulse to add convolution resonance is a plausible approach (but it is still linear!) and could be used to fake it for dry or overedited samples, but can't be as good as the original recorded resonance.

I'm still wondering why we think HW needs an enhancement to support pipe/case/chamber resonance, unless it's been removed from the samples. I can see the need when samples don't retain whatever resonance was recorded, and there's a desire to model the effect when the real thing was discarded. But otherwise HW does what is needed, as far as I can tell.

Since resonance is audible, it's already recorded in the raw per pipe samples. (Let's assume the release isn't shortened to cut off resonance too early and any denoising only kills background noises.

(Snip)

But pipes don't decay while sounding, and organs are pretty much a fixed resonance configuration, not responding differently each time the same pipe is played. Recorded samples and linear superposition should do the job. So why aren't we hearing it? Maybe denoising and release truncation on dry/close samples? Yes, getting rid of ambient noise like HVAC, traffic, blower, and humans is really hard, so maybe keeping resonance in the samples isn't even feasible. And maybe resonance is drowned out by reverb on wet samples?

Using an impulse to add convolution resonance is a plausible approach (but it is still linear!) and could be used to fake it for dry or overedited samples, but can't be as good as the original recorded resonance.

What I’m saying is that there is a temptation/tendency to overdo NR on a dry sample (they’re much more forgiving). Adding resonance back in via impulse is a fix after the fact, not something that should have to be done if the noise reduction is done more carefully.

The thing is, in smaller registrations/less dense texture you don’t notice it as much… it’s when you have a lot of stops and notes going that you feel like some warmth or resonance is lacking.

I’ll be curious to apply the lessons I’ve learned from Blackburn when I de-noise the big UMN Aeolian-Skinner… it’s max 1.5~2 sec reverb, even in such a large room, so a bit drier even than IPLA. I suspect I’m going to get results that are more true to life and resonant.