VIRTUAL ITALIAN ORGANS
by Andrea Bonzi
Among the various factors that contribute to the realism of a sample set for Hauptwerk, the Tremulant effect is certainly one of the most important - and also one of the least easy to reproduce: beyond the fact that even in pipe organs there are more or less successful Tremulants (we believe that the sense of annoyance given by an improperly regulated Tremulant or by random functioning is a common experience for most organists ...) it is very difficult to reproduce in a convincing way the interaction between the periodic fluctuations of flow rate and pressure of the air and the sound of the pipes, with sometimes noticeable differences between one register and another.
Hauptwerk offers the developer of the sample set - and therefore the user - three possibilities for the implementation of the Tremulant, each of which has its strengths and weaknesses: curiously, almost all producers have concentrated on the first two, leaving out the third, probably generally considered too complex and unreliable.
The first method of implementation consists in exploiting the Tremulant modeling integrated in Hauptwerk.
The system is based on particular very low frequency samples, whose waveform exactly reproduces the sound fluctuations produced in each pipe by the Tremulant: these samples are obtained via software by difference between pairs of identical recordings for each single pipe (or more often for a significant percentage: for example two samples per octave) one of which is taken with the Tremulant on.
By operating the Tremulant in Hauptwerk, the waveform of these samples is applied to the sounds, credibly simulating their effect.
Of course the achievable level of realism strongly depends on the number of samples used to obtain the difference: we have given the example of two samples per octave, but ideally you should resample every single sound.
One can easily imagine how this makes sampling sessions much more expensive, which can double for each register affected by the Tremulant.
A second possibility consists in the use of real samples recorded with the Tremulant engaged: a possibility often preferred by those producers wishing to reproduce with the maximum realism Tremulants with a particularly good effect, which the integrated model of Hauptwerk can only approximate without fully rendering their nuances.
Faced with a doubling of the time needed to sample the registers involved, the most faithful modeling of the effect as found in the original instrument is obtained, at least in theory.
In fact, this solution is the one that presents the greatest disadvantages by far.
In the first place, the RAM request strongly increases, making it necessary to load a double series of samples for each register concerned; and with this system it is virtually impossible to reproduce the Tremulant's valve start and stop correctly - the two samples are simply swapped out for each other with an unnatural "on / off" effect.
A third problem, subtle but equally substantial, concerns the impossibility of having strictly phased fluctuations between one sound and another. In acquiring the series of samples with the Tremulant inserted it is practically impossible to make each sample attack exactly in the same point relative to the oscillation of the valve: one will attack at the maximum point, the other at the minimum point, a third in an intermediate point ... The consequence, even if at first sight not immediately audible, is a phase shift between the various samples that will tend to debase the overall effect of the Tremulant making it more indistinct and in any case different from what can be heard in the instrument real, where the pulsations are strictly in phase for each sound.
A third solution involves the implementation of the Tremulant using the Hauptwerk wind model: it involves inserting, within the modeling of the bellows, a reproduction of the characteristic valve of this accessory.
As we said at the beginning, this system has practically never been considered by the main producers of sample sets for Hauptwerk: we believe that one of the reasons is the objective difficulty in controlling its parameters, especially where the modeling of bellows and wind chests is particularly complex. On the other hand, the developers of Hauptwerk themselves seem to advise against its use, giving preference to the other two methods.
We have recently started studying this last method, doing some preliminary tests that led to the implementation of a Tremulant obtained through the wind model in our sample set of the positive Ott 1972. The results achieved, although still partly experimental, lend themselves to some consideration..
The Tremulant with the wind model, although in fact not very easy to manage during the compilation phase of the ODF, has some interesting advantages: first of all, by acting on the air supply exactly as in the pipe instrument, it is able to reproduce - if one has the foresight to carefully calibrate the response of the "virtual pipes" to the wind model - in a very credible way the changes in frequency and intensity induced in each single rod.
With this system, as with the integrated model discussed in the first point, the problem of the characteristic phase of the model with double samples is eliminated at the root; compared to Hauptwerk's native Tremulant model, however, the dynamic nuances and frequency oscillations that play so much part in the "rendering" of the effect seem to be better reproduced.
A third particularly interesting point is the possibility - successfully tested in the sample set of the positive Ott - of making the Tremulant adjustable by the user through appropriate controls in the graphic interface.
The regulation takes place by varying the proportion between the air entering and leaving the Tremulant and allows to vary both the frequency and the depth of the pulsations. A smaller amount of incoming air produces lighter pulsations while a greater amount increases its depth; leaving the quantity of incoming air still, the variation in the quantity of outgoing air influences the speed of the pulsations.
For a better exploitation of this possibility, the sample set of the positive Ott has been equipped with four programmable memories in which as many positions of the two sliders that regulate the inlet and outlet air can be set.
As a practical example we propose here the explanatory video shown on the presentation page of the sample set, which effectively displays all the possibilities of the Tremulant modeled with this system:
The results obtained seem very promising and push us to continue our studies in this direction, in view of the implementation of this system in all our upcoming releases.