by Colin Pykett
here: 1 November 2010
Last revised: 20
Copyright © C E Pykett
This article was commissioned
by Organists' Review and published in the November 2009 issue. It
updated a previous one published some years ago in the same journal . It
discusses the pros and cons of the main types of digital organs currently
available, in particular the newer techniques of physical modelling and
so-called virtual pipe organs. Neither were part of the digital organ
scene when the earlier article was written.
on the section headings below to access them)
of digital organ and how they work
and cons - which method is best?
Digital organs continue to arouse controversy to the extent that a minority declines to call them organs at all. Yet they are otherwise well accepted in musical life, given that they inhabit buildings of all shapes and sizes from cathedrals and concert halls downwards and are played by professionals of the highest calibre. Therefore this article examines the technical situation as it is today with the intention of improving awareness of a field which continues to evolve. This is not meant to patronise the reader but the subject is rather complex, and customers for all hi-tech products stand a better chance of getting what they want if they are educated in the matter of buying good ones. This article updates an earlier one which appeared some years ago in Organists' Review
. Issues highlighted this time include the technique of physical modelling synthesis and the appearance of so-called virtual pipe organs, neither of which was part of the digital organ scene when the earlier article was written.
Types of digital organ and how they work
Three sound synthesis techniques are used in current digital organs - sampled sounds, additive synthesis and physical modelling. Each will now be described with reference to the digital hardware and software used in current instruments (often called 'sound engines' in the digital music industry).
Sampled sound synthesis
A sampled sound organ contains a large number of pipe sounds which will often have been recorded (i.e. 'sampled') on a pipe organ. The samples are stored as waveforms in the memory of a microprocessor system, which also enables them to be quickly accessed and passed to the loudspeakers as stops are selected and notes played. This is the oldest and still the most widely used technique in digital organs, having been introduced by Allen about forty years ago. However modern sampled sound synthesis is vastly superior to the early implementations because of the enormous progress in microprocessor technology over this period.
Figure 1 shows two versions of the waveform of an organ pipe stored in a computer. This sample resides in a digital organ system which I developed myself, and I originally recorded it on the beautiful Rushworth and Dreaper organ at Malvern Priory. In the pictures time runs horizontally from left to right, and the voltage sent to the loudspeakers is represented on the vertical axis. The duration of the sample is about eleven seconds, therefore it is not possible to discern the large number of individual waveform cycles in Figure 1(a) which shows the entire captured waveform merely as an amplitude envelope varying over time. However one can discern the three main parts of the waveform - firstly the attack transient which is heard briefly as the pipe settles down to stable speech, followed by the sustain or steady state phase while the key was held down during the recording, and finally the release transient when the key was released. In the diagram the word 'steady' is qualified by quotes because the steady state is not quite invariant. It exhibits small fluctuations in amplitude and frequency which endow the sound with the liveliness we associate with a real organ pipe, and this unsteadiness can be seen in the picture - the envelope was somewhat ragged rather than flat while the pipe sounded. There are also other artefacts such as wind noise. Thus the so-called steady state speaking regime is only comparatively steadier than the rapid changes which take place during the transient phases.
Figure 1(b) shows an expanded version of the attack transient in which individual cycles of the wave can be seen.
Figure 1. Sampled waveform of a pipe sound stored in a digital organ
Returning to Figure 1(a), sampled sound organs work as follows. When a key is pressed with a stop drawn, the corresponding waveform is rapidly selected by the computer and its attack transient is emitted from the loudspeakers. Provided the key remains pressed, the steady state sound is next read out from the memory until the loop end marker is reached (several seconds later in this example). At this point the system instantaneously jumps back to the loop start marker and the sound will therefore continue indefinitely as long as the key is held down. The loop points are selected carefully so that audible discontinuities are minimised. Nevertheless it is desirable to use samples of a length such that looping will seldom be invoked in normal playing; this was the case here where the sample was more than ten seconds in duration. However not all sound engines in commercial organs enable this to be done because of limits on the amount of memory which they are able to use.
When the key is released the system jumps immediately to the release transient to the right of the loop end marker, and ideally the sound then decays naturally. Like looping, this is another tricky aspect of sampled sound synthesis because there is no way of knowing in advance the exact point on the waveform at which the key will be released. Consequently it can be difficult to guarantee an undetectable 'join' between this point and the start of the release transient, and some manufacturers are better at achieving this than others.
In additive synthesis the waveforms are not stored directly. Instead, lists of numbers specifying the amplitudes of each harmonic are stored and they are then used to build up the waveforms in real time as required. The method was first applied to digital organs in about 1980 by Bradford university.
Figure 2 illustrates the general idea of additive synthesis by showing how a square wave can be built from a series of odd-numbered harmonics with the appropriate amplitudes. Only three harmonics are shown but the square wave (marked "sum") can be seen emerging as they are added together. Any periodic (cyclically repeating) waveform can be synthesised additively in like manner from its individual harmonics. The number of harmonics required for accurate tonal synthesis varies widely - hundreds can be necessary for the lowest notes of a pedal reed but less than ten for a Bourdon.
Figure 2. Illustrating additive synthesis
It can be difficult to simulate the attack and release transients using additive synthesis because it is a challenge to synthesise a rapidly changing sound using a finite number of harmonics, indeed the attempt can be mathematically meaningless. Similar theoretical constraints also limit the fidelity with which artefacts such as wind noise can be synthesised in this type of instrument. This differs from a sampled sound organ which simply replays the pre-recorded waveforms of actual pipes which have been put into it including transients, wind noise and all. For this reason some instruments using additive synthesis are forced to also incorporate snippets of sampled sound to cater for these difficult aspects of the sound being simulated
. Therefore additive synthesis alone does not always reproduce exactly what you might hear from a real organ pipe because in general it cannot, even in theory.
Physical modelling synthesis
Recently, digital organs have appeared which use a newer technique called physical modelling, although it has been long established in the wider synthesiser world. With the two older methods just described the organ sound engine recreates predefined sounds without any consideration of how real organ pipes actually work. Thus these instruments have no internal representation of acoustical physics beyond the numbers describing the waveforms or the harmonic structure, and until the numbers are loaded into its memory at the factory a conventional digital organ cannot make the merest squeak. At that stage it is analogous to a pipe organ without any pipes standing on its soundboards. With physical modelling, sounds are not recreated from stored waveforms or harmonics - they are created from scratch by mathematical models that ultimately produce those waveforms. Although at first sight the difference might seem academic, one needs to understand what it means if one is to understand what physical modelling is about. Thus one is now defining a model of an organ pipe in the form of equations, and one programs the equations into the computer rather than putting in the waveform of the pipe or its harmonic structure. Then, when one presses a key, this sets the model running and it will duly calculate in real time the waveform of the pipe. The waveform does not exist until this point, just as the sound of a real pipe does not exist until one admits wind to it when playing a pipe organ.
What advantages does this confer? The answer to this question is easy when modelling the sound of a piano, for instance, but less clear for a pipe organ. This is because of the relatively invariant nature of organ sounds compared to those of a piano. Not only does a piano note change rapidly as it sounds but it is different each time, whereas for the organ these differences are much smaller. The harder you strike a piano key the louder the sound and the tone quality varies as well, as do the attack transient and the decay characteristics. The differences are subtle and important, thus simulating a piano properly is far more difficult than simulating an organ. But by setting up a mathematical model of the piano action for a key, its strings and the soundboard, the computer is able to calculate the sound waveform corresponding to any key velocity, which was difficult until the advent of physical modelling. However, organs are different because the almost invariant nature of pipe sounds can already be reproduced more or less exactly by current digital techniques. Thus a digital organ using physical modelling cannot offer the major advantage just outlined for the piano because it is not called for with the organ, so it is difficult to see it as the major capability leap that it has been for other instruments. It could certainly simulate the relatively minor variations in pipe speech discussed already, but most, if not all of these, can already be simulated using existing synthesis methods.
Space limitations preclude a more detailed discussion of physical modelling here but an extensive article is available elsewhere
Pros and cons - which method is best?
When done properly all three synthesis techniques can render convincing simulations of organ sounds. Other areas such as loudspeakers are often more important and these will be discussed presently. Nevertheless, there are some differences in synthesis capabilities which have already been mentioned. Only sampled sound synthesis captures more or less exactly the subtle sound of an actual organ pipe while it speaks; the other two methods can only approximate to it via various mathematical processes. Therefore the fidelity of details such as attack transients, wind noise and random fluctuations in frequency and amplitude during the steady state phase of pipe speech will in general be best for sampled sound synthesis. For this reason, if asked to state a preference and all other things being equal, I opt for the sampled sound method for what it is worth. I have come to this view after some thirty years spent analysing the sound of organ pipes and trying to reproduce them digitally.
But, like so many other things in life, it is horses for courses as it is with pipe organs. Some people like Compton's organs whereas others would rather die than listen to anything but a Silbermann. Therefore perhaps we should be content to reflect on the happy circumstance that digital technology has reached a level of capability which now offers us a choice of three methods, none of which is perfect but all of which can deliver good results in principle.
Virtual pipe organs
Over the last decade a type of digital organ has appeared which has become known as the virtual pipe organ (VPO). This is an unfortunate name because if an organ is virtual it is plainly not a pipe organ. But given that the name is widely used, it is important to realise that VPOs do not use yet another form of sound synthesis. They use the sampled sound method already described and it is only the implementation which differs.
The main difference is that VPOs use one or more desktop computers instead of the special purpose sound engines traditionally used in sampled sound organs. Now that they are so fast, and have such enormous memories, it is possible for a computer to hold enough sound samples for every note of every stop, thus the number of samples can equal the number of pipes in a pipe organ with the same stop list. Such sample sets copied from pipe organs, pipe by pipe, are used by some VPOs and thus arose the name. Together with the software which the originator of the VPO provides, one can connect a standard console using MIDI to the computer and play a range of different simulated organs in one's home or elsewhere.
Some large instruments have been installed using this technology. At the other end of the scale there is a low-cost niche market largely populated by enthusiasts who enjoy DIY electronics and/or have above-average computer expertise (these remarks are not intended pejoratively because some are highly competent). However there are some downsides relating to the latter products. Both electronics and computer know-how are desirable because backup and support scarcely exist for most of these systems. And, although the startup costs might seem small on the face of it, provided one already has the PC, it should be borne in mind that one needs a console and a more or less elaborate amplifier and loudspeaker installation. These can be expensive, and interfacing them to a computer is often a non-trivial undertaking even for those with experience.
There are also a number of irritations, particularly with the low-end systems. Not the least of these is that one has to wait for the PC to boot up before the instrument can be played, and in the case of large sample sets this can take several minutes. Sometimes the monitor has to be visible which can be intrusive, and some systems require the player to frequently interact with it via the mouse or keyboard. The cooling fans in many PCs are also unacceptably noisy in a room where music is to be played. But if one is prepared for all this, then some VPOs can offer excellent sound quality when used with a good sample set.
Other technical issues
The previous article already referred to  touched on many technical issues and to avoid repetition most of them will not be raised again. But in view of changes in technology it is worth revisiting some topics.
Number of samples
A rank of organ pipes exhibits both systematic and random changes in volume and tone quality across the keyboard and any type of digital organ must simulate this. The obvious goal for a sampled sound instrument is to incorporate a separate sample for every note of every stop, and some of the sample sets available for PC-based virtual pipe organs do just that. Although some of the more traditional instruments also offer this feature, others do not, and in these cases each sample is 'stretched' across a range of notes. Today it would be difficult to excuse a manufacturer which did not use at least four samples per octave for every stop (one every three notes). It would suggest that the hardware and software synthesis environment on which his organs were hosted might be of limited capability and out of date. As for additive synthesis instruments, it seems that even some recent systems might have difficulty with this also
. With physical modelling the situation is different because samples as such are not used. However, it is still valid and necessary to enquire whether the organ in question is able to simulate sounds independently on a note by note basis for every stop.
This term is used here in a digital music context rather than referring to Palestrina or counterpoint. For digital organs it means the maximum number of simulated pipes which can sound simultaneously. All types of digital organ use a finite number of generators or oscillators (the exact term varies from one maker to another) to actually produce the sounds, although these have nothing to do with the tone generators and oscillators formerly used in analogue organs. For sampled sound or additive synthesis instruments each generator is a flexible digital circuit arrangement, or a fast software module, which can accept any waveform for any stop and route it to any amplifier and loudspeaker. When a key is pressed the computer determines which waveform has to be used for a particular stop and assigns it to a currently free generator. With physical modelling each generator is a software module which computes the required waveform.
Early digital organs only had a small number of generators and thus were of excessively limited polyphony. They would therefore tend to 'run out of notes' when chords were played on large numbers of stops. Although this problem is less likely today, you should bear in mind that super-octave and sub-octave couplers dramatically increase the polyphony load on any organ sound engine. Therefore if you are confronted with an organ which contains such couplers (many do not precisely for this reason!) you should test it particularly thoroughly for the missing note problem. It is not difficult to calculate the required polyphony figure for a given stop list, and you should seek independent advice if necessary before committing to a purchase in view of the importance of the matter.
The importance of loudspeakers cannot be overestimated. Among many other factors it is a well known phenomenon with digital organs that, while individual stops or small combinations might be good or excellent, anything beyond this can become progressively more unconvincing and sometimes unpleasant. Two reasons for this are distortions due to intermodulation and those caused by mixing too many signals into too few loudspeaker channels. These matters are too involved to explain here but I have addressed them in detail elsewhere
. The consequence is that considerably more loudspeaker channels are required than manufacturers often provide, with at least 24 being a reasonable goal to aim for. But this means a large and expensive installation, of course.
It is impossible to address here all the requirements for a satisfactory church or concert hall installation other than to emphasise the importance of getting it right. It will often be worthwhile retaining the services of an independent consultant to advise on the proposals made by the manufacturer.
Although digital organs have come a long way since they first appeared around four decades ago, it remains a fact that they still suffer from a number of intrinsic defects which have dogged them all along. Therefore it is also a fact that pipe organs retain the edge in terms of sound quality today because (apart from extension organs) they have no limits analogous to polyphony, the number of inbuilt sound samples and other factors which plague digital instruments, and equally importantly, nor do they suffer at all from distortion. Consequently there can be little argument that a good pipe organ, though not an indifferent one, should be the instrument of choice when space and funds permit, at least in a public building.
However, in the considerable number of cases in which space or funds do not permit, then a good digital organ, though not an indifferent one, can sometimes offer a cost-effective alternative; but the cost and space saving might be less than first envisaged if the necessary and vital attention is directed towards the loudspeaker installation. Moreover a high quality console built to pipe organ standards is also expensive in itself, and they are not always found in the standard offerings of many manufacturers. These two factors alone - console and loudspeakers - will often dwarf the cost of the digital sound engine employed, although it will still be necessary to ensure that the correct one is chosen because it can make or mar the final effect. Hopefully this article will assist in selecting it.
1. "Electronic Organs", C E Pykett, Organists' Review, August 1998.
Available on this website (read) .
2. "Music's measure: using digital synthesis to create instrument tone", P and L Comerford,
Organists' Review, May 2007.
3. "Physical Modelling in Digital Organs", C E Pykett, 2009. Available on this website (read)
4. See the section entitled "Technical Matters" in the article on this
website "The End of the Pipe Organ?" (read).
Colin Pykett has played the organ since his schooldays. Following a PhD from King's College London he pursued a career as a physicist and he has also continued an active involvement with electronic organ technology over some forty years, though he has no connections with the industry. In addition he has carried out detailed research into sound production by organ pipes as well as several aspects of pipe organ mechanism, and he has undertaken particular studies of the technical contributions of Robert Hope-Jones to these areas. All this activity is documented in numerous contributions to the literature over many years and on his website at