VIRTUAL ITALIAN ORGANS FOR HAUPTWERK

Ambient Acoustics:
Multiple Releases or Convolution Reverb?

by Andrea Bonzi

One of the elements that distinguishes Hauptwerk decisively from other similar software is the way ambient acoustics are managed - that is, the reverberation of the spaces, generally churches, where the reproduced organs are located.
Early users will remember that a big limitation of HW1 - Hauptwerk Version One - in the face of unprecedented realism in the sound rendering at steady state and above all of the transient attack as well as the stereophonic spatiality (everything singularly depended on the quality of the audio recording during sampling) was its inability to make the reverb tail correctly in relation to the duration of the notes: even a "very detached" response of the ambience to the entire sound at steady state, often lasting many seconds, was inexorably reproduced with an unnatural "bell effect".

A first attempt to overcome this inconvenience was to give Hauptwerk the opportunity to read samples where the sound detachment could be set at two different points, in order to partially cut the reverb tail. A compromise solution, still far from reality, but yet, it was an improvement. Already from HW2 - Version Two - another approach was used, which consisted in sampling the reverberation tails separately using different times, from the shortest staccato, to the sustained, to the very long note, leaving the task of mixing them to the steady-state sound according to the duration of the notes, to the software.

This is the system still in use today in version 4.00 and later: in addition to multiple reverberation tails, there is the possibility of implementing more attacks for each single note, thus being able to obtain a very credible representation of the small variations found in the real instruments - at the cost however, of a very remarkable RAM memory consumption, since each individual sound normally presupposes at least three or more different samples and calculating that for each different attack a sample of the entire duration of the sound at steady state is required.

Nonetheless, if during the sampling phase the durations of the release are carefully chosen and if the same care is taken in calibrating the definition file during the production of the set of samples, the final effect - also thanks to the power and flexibility of the algorithm with which Hauptwerk, especially in the most recent versions, manages the mixing - it is surprisingly realistic and can come incredibly close to the original acoustics of the ambient.

It is the opinion of the writer that, if you want a precise replica of a given pipe organ, this is absolutely the best system so far, as it requires a considerable expenditure of time during the audio recording and processing of the set and a 'equally large RAM memory demand during playback: for the simple reason that it is the only method to guarantee a sound image of the environment, not only real but also perfectly consistent with the direct sound coming from the pipes.

The Impulse Convolution Reverb

Next to sets of samples called "wet" (Anglo-Saxon) - that is where the acoustics occupy a good percentage of the single sound, forcing the use of the sampling system just described - there are others called "dry", in which - deliberately - we try to pick up a sound, in fact as "dry" as possible, carefully avoiding the ambient acoustics. In this case, the reverberation is reconstructed a posteriori and there are several software solutions that allow you to do this even with versions 3, 4 and later which provide for the use of Hauptwerk as a VST plugin.

The immediate advantages of this solution are evident: first of all the reverberation tail is exactly corresponding to the length of the note played and, not secondary, the consumption of RAM memory is drastically reduced both because it is sufficient to load a single sample per note (or all, at most those corresponding to multiple attacks) and because the samples themselves, not containing reverberation or almost, are much shorter in duration.

The downside is that, even with the use of excellent algorithms, it is almost impossible to reproduce exactly the characteristics of the original ambient acoustics with all its nuances and the articulated frequency response.

Impulse convolution reverberation is based on the principle of adding to the input sound signal a further signal that reproduces the response of a given environment to a single sound impulse. The principle itself is simple: an instantaneous signal, which covers all audible frequencies, such as a blank pistol shot or a burst of some kind, will produce an acoustic response of the ambient that can be recorded and applied through appropriate algorithms to the audio input signal.

If this is simple in principle, it's not at all simple in its practical realization, since it requires, in addition to first-rate microphone equipment, above all an audio apparatus capable of reproducing, with a sound power appropriate to the vastness of a church also of large size, an almost instantaneous impulse that covers at least the entire range of audible frequencies - or rather, the entire range that can be picked up by microphones. And even in this case it is very difficult to be sure of reproducing exactly the acoustic response of a given environment.

Another category of problems arises when the impulse response thus obtained is applied to the individual samples: this is evident considering that each sample corresponds to a pipe placed in a precise position in the environment, the response for which will be different from another post in a different position and in any case different from that of the source used for the impulse - obviously it is not conceivable to have impulse sources to be positioned at each individual pipe inside the instrument, recording each time the different response, also because there is no software capable of managing an impulse response for each individual sample.

Furthermore, since almost all sampling takes place by means of stereo capture, problems arise in the coherence between the stereophonic image (taken at a precise distance, fixed during sampling and therefore not modifiable and usually short if it is a "dry" sample) and the variation of the relationship between direct and reverberated sound that each algorithm allows: we can move the sound image at will by increasing or decreasing the proportion between direct and reverberated sound but the stereophonic image - unless that it was taken using the mid-side technique and where the algorithm in use allows it to be decoded in real time, with variation of the stereophonic amplitude - remains unchanged.

In this way the ear receives contradictory messages: at a great distance suggested by the wide reverberation, a wide and very detailed stereophonic image contrasts, which instead suggests a proximity to the sound body: from this contradiction arises the feeling of unnaturalness that sometimes accompanies the use of impulse convolution reverberation, unless you find some system to compress the stereophonic (not reverberated) image - a relatively easy thing to do in postproduction but certainly not in real time.

For these and other reasons the developers of Hauptwerk - and with them a large part of the users - have always considered the method of multiple releases as superior which, however compromising, at least guarantees absolute fidelity in the correct return of the ambient response to the sound of each sample.

Hauptwerk V and the Integrated Convolution Reverb

Nonetheless, in the very recent release of Hauptwerk Version Five, the possibility of using an integrated impulse convolution reverb is included, which finds its strength (from what you can read on the net) in a new management of the audio mixer, in particular its possibility of being applied independently, if not exactly to each individual sample, to a maximum of 1024 channels to which groups of samples can be addressed.

Some predefined impulse responses are integrated into Hauptwerk in a proprietary four-channel format: it is not clear whether the user can add his own.

This is undoubtedly interesting and makes the use of "dry" sample sets extremely flexible; however, it changes direction from the notion of accurate modelling that includes the ambient acoustics of a given instrument, an idea that has been - and still is - one of the strengths of the launch of hundreds of sample sets by many manufacturers.

Our opinion is that if you want a set of samples to give a realistic feeling when being played, you need to pay close attention to any kind of forcing. If this is true as regards the use of the "extensions" as Hauptwerk's definition file compilation allows, it is more forcefully true for a much more delicate thing such as the rendering of ambient acoustics.

The use of a convolution reverb can be fun, and you can indulge yourself in listening to a given instrument "as it would sound if ..." it was placed in a fictitious ambient with characteristics corresponding to that acoustic. But in our opinion, it must be clear that these are precisely fictitious operations: the musician's ear must first of all judge their validity.

Given all these considerations and given that the objective of our production of sample sets for Hauptwerk is mainly the faithful sound documentation of the instruments we sample, without applying force, the virtual organs presented in these pages willnormally include multiple releases as always practiced so far by all manufacturers according to the most valid universally recognized system; "dry" sets or sets with little reverberation will be released occasionally to which the new convolution reverb integrated in Hauptwerk V may be applied.

We do not recommend applying reverb to sets where the original environmental acoustics are already present: Hauptwerk allows you to load them by cutting the reverberation tails, but the acoustics are also present in the steady state sound and would overlap the added reverberation with unnatural results.