cymbalom (Page 2 of 2)

Modartt is the company; Pianoteq is (just) a product.  Pianoteq *could* be modularized as The Dragon says. Should it be?  I doubt it.  I think there is danger in trying to make one product be all things to all people.  I would however like to recreate the acoustic double bass that was available in PTQ 2 as I have the score for Miles Davis' Flamenco Sketches and already have a physically-modeled trumpet.

I have been extensively involved with Hauptwerk for 5 years, beta-testing version 2 and involved in production of several sample sets on various levels.  There are two "factions" in the Hauptwerk community; those who are striving to produce accurate facsimiles of existing pipe organs (including their natural acoustics) for home use/archival purposes, and those (fewer in number and considerably less commercially vocal) wishing to support and augment existing pipe organs with digital technology and provide viable alternatives to expensive pipe installations.  It is the latter camp to which I (mostly) belonged.

It is true that Hauptwerk relies on multi-gigabyte sample sets but unlike conventional samplers and due to the nature of the organ (anything can be sounded at any time and in any combination) the entire organ must be loaded into RAM.  Just think about that for a second.  It just isn't practical at the present time to sample a large organ with samples longer than 10-15 seconds, particularly at more than 44.1/16 bit. 

It is correct that organ tone does not decay, but conversely the tone is far from static and a looped sample is hardly optimal in this application.  Hauptwerk does (where possible) use multiple loops within samples to make the result less boring.  Looping these samples is no mean feat, particularly if the recording is up-close rather than ambient.  Even assuming that all these conditions are met and a "perfect" set of samples is obtained, rather like the situation with sample pianos, you're left with a very narrow representation of a particular instrument which, even with Hauptwerk's built-in, real-time voicing controls (tone filters, nothing more, nothing less) is always going to have the same basic flavour.  You're not going to be able to turn a Viola da Gamba into a leathered Open Diapason No. 1.

(Please forgive the length; I've had a few)

One poster commented that there isn't much percussive attack on an organ.  Sure.  But there *is* an attack which is very complex and the nature of it is rooted in physics and the physical construction of the pipe/style of voicing.  Something which could be done with Physical Modeling that cannot be achieved any other way, would be akin to the hammer hardness and soundboard controls in Pianoteq; for an organ, (simplifying) the variables could be:

1) Diameter of pipe (narrow pipes develop more overtones than broad ones)
2) Ratio of mouth width to pipe diameter.
3) Height of upper lip of pipe (lower develops more overtones and quicker attack, generally a higher lip is used for orchestral flutes on copious wind)
4) Windpressure (complex interactions with other parameters, naturally!)
5) Wind flow rate (toe hole size, position of languid)
6) Taper of pipe (suppression/reinforcement of certain harmonics)
7) Stopped or open (pitch and timbre change)
8) Chimmneys (partially stopped) - development of higher harmonics - reedy tone

9) Scaling - timbre changes over the compass of the keyboard.  Ralph Downes' book "Baroque Tricks" gets into this heavily and his experiments at the Royal Festival Hall.

...and that's just for flue pipes.

To clarify, the Hauptwerk wind model modulates both tone and frequency of the samples in real time to simulate the (sometimes excessive) wind demand when lots of pipes are sounding.  While it *is* possible to radically change the tone of the samples by stating "pipe X is on 6" of wind" and then only supplying it with, say, 2", this will radically affect the pitch in unpredicable ways and wasn't the actual point of the model.  The actual tone generation in the engine is purely sample-based.

It is my dream to be able to, for instance, extend the compass of a stop on a real organ, either upward or downward, by applying real measurements such as those mentioned above, to a virtualized model.  Or to be able to replace one type of solo flute with one of a different character, without resorting to crude tone controls (ignoring time-domain features) or swapping out entire sets of laboriously-created samples.  Never underestimate how laborious a process sample editing is.

What a great post, Neil!

Question: seeing how well WIVI did well for wind instruments, and with such a low CPU usage too, don't you think that Arne Wallander is on to something there? After all, organs ARE keyed wind instruments!

I bet if Modartt and Wallander merged, we'd be in bliss!

And if QuikQuak then joined them, we'd have a draw-your-own-room-for-reverb in Pianoteq, then place tehe piano anywhere in that virtual room! Man, I really LOVE RaySpace reverb!

Au contraire, sir.  I think it would be a fantastic product.  What I meant was, I didn't think Pianoteq itself - particularly the interface - should be hacked up to do it. 

Regarding stability of sound, or lack thereof, yes there are pitch/phase interactions between pipes (which can be aggravated by looping of course) but I was actually thinking of the end result of exciting a column of air with a turbulent air-sheet.  Something like a pedal string pipe has a very slow attack, pronounced transient noise and a "burble" because the majority of the air-sheet is outside the pipe, due to how the mouthparts are arranged. 

Even a large-scale flute won't speak the same way twice and the character of the "random" air noise is a big component of the sound.  Of course the air noise isn't random, it's effectively filtered by the construction details of the pipe which I've already discussed, being set to emphasise odd-order harmonics for a stoppered pipe, for instance.

Pipe width affects tone because the natural harmonic resonances of the pipe resonate when the pipe is excited by the air sheet and depending on how the airsheet is angled, different "forced" harmonics are developed.

The natural resonances are *not* exact multiples of the fundamental, becoming progressively sharper higher up the series.  This is due to end-correction; the pressure waves traveling up the pipe overshoot the end, so the pipe appears to be longer than it actually is.  This correction is more pronounced for the fundamental and low-order harmonics (which sound relatively flatter than the should be) and is more pronounced for wider pipes.  What this means is that for a wide pipe, there is more disparity between a plot of the frequencies of the natural harmonics compared with the forced harmonics.  This results in less amplification of the high-order harmonics (by resonance) and a duller tone.

An easy experiment to show the effect of the airsheet is to blow across the top of a beer bottle.  Varying the angle of your lips (the airsheet) will vary the tone from stringy to fatter.  Couple this with a thinner or fatter pipe  and even this crude arrangement can yield a variety of tones.  Now support the bottle on a table and place your hands vertically at the corner of your mouth, making a seal with the bottle top.  Notice that this concentrates the airflow, making the tone much fuller and the speech more prompt.  Exactly this arrangement is used on many pipes which have "beards" of metal soldered onto them.

oh yes, +1!

Okay, extract of Percy Whitlock's "Dolcezza" uploaded.  Bear in mind this is additive synthesis.  The noise components are intended to simulate the interaction between the different frequencies of the natural and forced harmonics.  It is very rough and in places the noise is completely overdone as I was trying to see what the limits were (not very high!) for proper looping.  Some notes in the last few chords, you'll hear the tremulous effect which is a high level of natural harmonics interacting with basically a similar output level of the forced lower partials.  I think I only properly sorted out a couple of notes in the lower octaves, the rest being mostly inter and extrapolated which only works to a point. 

The breath noise is bands of oscillating sine waves in groups, say partials 20-30 and a higher band of partials 50-80.  These meander around the spot frequencies in a semi-random fashion and the end result was designed to sound like air-sheet interaction.  Obviously in a pipe of this type, there is no excitation above about the 8th partial but this noise is vital for the "believability factor" and missing entirely from conventional additive synthesis, or in later incarnations of Bradford/Musicom, added in by using sampled noise.

There has *got* to be a better way of doing this than guesstimating/doing it by ear with additive synthesis though.

Agreed.  How are you getting on with your streched tuning data?  Do you still need someone to look at the Steinway M?

On the matter of the Pianoteq interface, I'm not suggesting hacking it to make an organ (a sacrilege), but a developing completely new software called Organteq.

I think the piano technician background of Phillipe Guillaume was critical in being able to develop Pianoteq.  A mathematician without this background would have a very difficult time understanding the interactions of the parts of a piano that affect the sound.

http://www-gmm.insa-toulouse.fr/~guillaum/cv_en.htm

Perhaps with your knowledge of organs, you could be in for a career change - math skills are learned.  Or you could learn French and collaborate.

Glenn

@ Glenn:  Yes, the PLAY-style interface is an idea but even for a small chamber/continuo organ (a nice proof of concept and I have quality, unedited samples of one for analysis purposes) the main interface would only have enough detail for setup of the entire instrument; temperament, acoustics, listener distance (markedly alters tone due to much of the noise not carrying far) and balance between ranks.  For the different ranks of pipes, there would need to be a - or several - deep editing modes.

@Jake:

Somehow I missed the rest of the discussion about the tuning data and Scala files.  In short, I don't know anything about Scala files but I'm quite prepared to learn if that's going to help someone.

In respect of fractions of cents, Philippe et al would have the definitive answer as to whether this matters to the PTQ engine.  If it doesn't, the probable reason would be that a struck piano string won't keep to a static frequency, unlike an organ. 

If you held down 2 organ keys an octave apart and you had a tuning discrepancy of less than 1 cent, you may well hear the phase shifts over time.  This would be more noticeable on an electronic, because of the ways that electrical signals interact and phase cancellations are much greater in magnitude than with pure acoustic mixing. 

Tuning organs is tricky business because in particular, reeds will produce enharmonic noise which can beat against mixtures etc.  I don't see this being a problem with piano because the sound is constantly evolving in spite of your tuning accuracy.

I love the organ too and sometimes play a sampled instrument from Sampletekk, but I would love a physical model.

Some people claim to be already doing it:

http://www.domusorgels.nl/en/unico/physicalmodeling

or are studying it:

http://www.sea-acustica.es/Sevilla02/mus06011.pdf

Has this produced any worthwhile result that you know of ?

Yes!

Yes, I can hear this too. I hear two things - a pleasant "click" right at beginning, and then I hear very high frequency, metallic overtones that decay rapidly. As you say, it really does sound like a steel string being struck.  It sounds *clear*, too - hi fidelity!
I haven't tried your layering idea yet.

Greg.