Luc Henrion wrote:While an organ needs just one (looped) sample for each note and each stop.
This is much to simplistic. Organs with that style of sampling or synthesis, sound completely "dead" compared with the real thing. With no disrespect to what appears to be a one-man operation (though the details are quite well hidden on the web site!) listen to the "virtual organ company" demos to hear how weedy and uninteresting that sounds compared with a real pipe organ. Maybe that sort of sound was "good enough" back in the 1960s or 70s, but not any more.
One obvious issue is that the initial transient speech of a pipe may take several seconds (literally) to become stable. Most notes are shorter than that, so as a minimum you need several different "release samples" to capture the ending transient of the notes. If you just fade out a note that is still "growing", the result sounds very wrong.
More subtly, the "transient" resonances of an organ pipe are very different from the spectrum of the steady tone - in fact the fundamental transient frequency can be a long way from the steady pitch of the pipe, and the different resonances are rarely in nice simple ratios like those of a piano string (ignoring the small amount of inharmonicity in the string, to keep the discussion simple). If you don't capture that correctly, you lose the most of the definition of the attack and release of each note, which quickly turns "articulation" into "a muddy sounding mess".
But the most important missing "secret sauce" is that the sound of the organ (like any wind instrument) depends on the stability of the wind supply throughout the duration of the note. No organ has a completely stable wind supply, which is hardly surprising when you consider that starting to play a single note at the bass end of a keyboard may consume more wind on its own than a 10-note chord of higher pitches!
The blowing system is designed with a lot of mechanical devices to try to compensate for this, but because of friction and the impossibility of "predicting the future" they don't work "perfectly". In fact, if there are several of them connected to the same part of the wind supply, they can start interacting with each other to set up small oscillations in the supply, until the organist happens to play a note that stops that happening. If you watch the behaviour of these gizmos with the outer case of the organ removed, the total visual effect is more like watching open-heart surgery in progress than looking at a "simple" mechanical device operating.
Physical modelling of an organ without going to that level of detail isn't going to give very good results - but then neither does recording an organ in any case. You can't fit the huge number of simultaneous independent sounds sources (pipes) being played - literally thousands, for big chords on full organ - into two stereo channels. The sound systems for large-scale electronic instruments have many more independent amplifiers and speakers - sometimes between 50 and 100, not just two!
Luc Henrion wrote:If you want to model an organ, there should be one model for each stop. Not realistic, CPU wise, when adding samples upon each other is so easy.
Having spent several years "playing" with this technology as a hobby (and I think I know enough from my "day job" to do that seriously, and not just try to re-invent the wheel starting from zero) I completely disagree. You need a separate model for each pipe, not just for each stop, if you are going to make any progress capturing the interactions between the sounds from a few hundred or thousand pipes enclosed in a relatively small space.
In physical modelled pianos, it is those interactions that make the difference between the sound you get from Pianoteq, and a sample set which will inevitably sound like "88 different pianos, each playing a single note" - a very different thing.
Actually, this is becoming realistic, CPU-wise - many of the pipe models will be similar except they are working on different data, and they quite well suited to run on the GPU of a high-powered graphics card. Within a few years, the cost of a dedicated computer to run this sort of calculation will probably be less than $1000 - cheaper than the current top-quality organ sample sets today.
But I thing the take-home message from all this is quite clear: not only is physical modelling a pipe organ is more complex than modelling a keyboard instrument, but also very little of what really matters to get good results is common between the two models.
Last edited by Rob Tuley (09-09-2017 23:54)