THE SOUNDFONT SAMPLES PROJECT OF SEPT/OCT 2010

For some weeks during September and October 2010 I worked hard and for many hours investigating how to make quality soundfonts using recorded samples.

Over the years I have made many recordings of pipe sounds in organs I had easy access to. My practice has been to record every second note. Most of the stops in Earlwood Organ No.1 and No.2 were based on those recordings, and in the case of those stops, the samples were synthesized using either Klangsynth or the later AdsynDX program. The samples so produced are a very accurate representation of the "average" harmonic structure of the steady-state sounds of the pipes used, at least within the limitations of the two programs (Klangsynth allows synthesis up to the 33rd harmonic, but AdsynDX allows many more harmonics). The actual analysis of the sounds was performed using Audacity. I might add that I far prefer to use AdsynDX, as I find it quicker to use, but even more importantly, it allows the use of spreadsheet techniques to handle the harmonic structure information "automatically", so that one hardly needs to get involved in all that detail. The writer of AdsynDX generously incorporated a file-import function (at my request) to make this possible. While not a free program, it is very inexpensive, and its file-import function means that countless hours of work are avoided.

I took the synthesis route because the quality of my recordings, in terms of motor/blower noise and air hiss, while not invalidating the harmonic structure information of the sounds, precluded their use as recorded samples. My recent work has convinced me that the SPEAR program has changed all that. There may of course be other programs that do the same thing, but SPEAR is the one I have become familiar with, and I certainly found it very easy and quick to use, and I believe it can remove unwanted components without degrading the final sound too much. (It would be unrealistic to hope that it might leave the sound absolutely untouched). I have in one post to the jOrgan Forum mentioned that I had found SPEAR to introduce (at quite low level) a slightly unpleasant quality in one test I made. I am now of the opinion that in that test I may have overdone my enthusiasm for deleting unwanted sounds, for I have not reproduced that effect in my recent work.

In that work I made use of recordings made of ranks in two pipe organs installed in the Sydney area in the early 20th Century. Pitt Street Congregational Church (now "Uniting Church") in the Sydney CBD is a large and well-known Sydney church. In 1910 they received what was the last pipe organ exported to New South Wales by the famous Hill and Sons of England, before they merged with Norman and Beard. The organ has three manuals, and its action is tubular-pneumatic. The other organ I used was built for the Katoomba Congregational Church (also now "Uniting") in 1915 by Charles Richardson, a major N.S.W. organ builder of that period, who learned his trade at his father's organ-building factory in England. It has two manuals, with tracker action, and is quite small. The recording quality of the latter organ's sounds was quite usable for recorded samples except for the motor/blower noise, but SPEAR was able to remove that noise. In the case of the larger organ, the problem with the recordings was the excessive "hiss" introduced by the capacitor microphone I was using. Because of the "unbalanced" lines involved, the recording system was vulnerable to the electromagnetic radiation present in the Sydney CBD, and although SPEAR could and did remove that, the resulting sounds were somewhat lacking in the levels of their higher harmonics. What was just as disappointing to me, was the attack characteristics of the pipe sounds, due I believe to a very old tubular-pneumatic action which was opening the windchest pallets fairly slowly.

So far I have made two soundfonts to demonstrate the results of this work. The first soundfont, PSU_Great_Stops_FS.sf2, is suitable for use with Fluidsynth, and contains samples based on the Open Diapason No.2 8', and the Principal 4' of the Hill organ at Pitt Street Uniting Church. Because of the problems mentioned in the previous paragraph, I have no enthusiasm for using these samples in later virtual organs. However, they are still of interest. The OD No.2 samples are completely synthesized (both attack and steady-state sections), but the Principal 4' samples are hybrid: recorded attacks with synthesized steady-state sections. In the case of the Katoomba soundfont, KU-Open_Diapason_8_FS.sf2 (also to be used with Fluidsynth), the samples are fully recorded except for two samples in the top octave. The looped sections are very short, involving only a few cycles of the waveform, and to make this workable, it was necessary to use SPEAR to remove any noise components from those sections. I did apply modest amounts of synthesized pipe air noise in the soundfont, to mask the complete absence of air noise from the steady-state sounds. I also applied modest amounts of the "Earlwood Tremulant" to the sounds so that their non-varying quality in the steady-state sections (unavoidable when using very short looped sections) could be masked.

I plan to make a number of presets of the Katoomba organ, using recorded samples, and to make them available on this website for any interested people to use. I plan also to incorporate them into further versions of the Earlwood Organ No.1, although this will not affect versions previously released.

Click HERE for the PSU_Great_Stops_FS.sf2

Click HERE for the KU_Open_Diapason_8_FS.sf2

NOVEMBER ADDITIONS

The next stage was to make a Principal 4’ preset to use with the Open Diapason 8’, based on recordings of the corresponding rank on the Katoomba Uniting Church organ. This time I chose to abandon the use of SPEAR to remove the motor/blower noise, and instead used High Pass filtering as made available by Audacity. I regard this as totally successful. I also chose to create hybrid samples, with the attack section recorded, but the steady-state section synthesized in order to make looping very easy. I regard this approach as in no way inferior to using recorded loops of short duration, as was the case with the Open Diapason 8’ preset. One advantage is that it allows very fine control over the pitch of the samples produced. With this Principal 4’ preset, I did not include use of the Earlwood Tremulant to provide “random” liveliness of the steady-state sound, as the stop would normally be used along with an 8’ stop, and so that provision becomes of less concern. There is no technical reason why it could not be provided.

I must admit that since identifying what I see as limitations in the SPEAR program (at least for the use we have in mind), I am less enthusiastic about using it. I think it may be used more "safely" in the case of some types of pipe tones than in others. I would be reluctant to use it where the noise content is of particular interest, or where the "random" fluctuations of individual harmonics are of interest, or where there is something particularly noteworthy about the rapid attack sound. I have decided to switch to the use of Audacity's filter functions, as a means of reducing the motor/blower noise problem. The point is that once you use SPEAR to process a sound file, even if it's only to remove the motor noise, the program has compromised certain other aspects of the sound, and this may be important to some hearers.

I have now made a combined soundfont of these two stops, which represents the main Great chorus of this small pipe organ. I regard the result as very pleasing indeed, and whatever one may feel about the actual tone of the stops, I believe that the method used, especially with the Principal 4’ preset, has been strongly vindicated. What this soundfont has brought to light (for me, at least), is the importance of tuning in gaining a satisfying effect. Playing these two stops either separately or together indicates the slight out-of-tuneness which is a clear characteristic of pipe organs. It is all too easy to forget this aspect when dealing with fully synthesized samples. Hearing the 8’ stop alerted me to its importance, and so when finishing the 4’ stop I included slight detuning, either in the samples or in the soundfont parameters.

Click HERE for the KU_Great_Stops_FS.sf2

This soundfont can be conveniently loaded into the Fluidsynth element of the Earlwood Organ No.1b, in which case the Stopped Diapason 8’ stop switch will activate the Open Diapason 8’ stop, and the Dulciana 8’ stop switch will activate the Principal 4’ stop. All other stop switches should be left OFF. Alternatively, any jOrgan disposition will do, as long as you know which stop switch activates Preset “0” (for Open Diapason 8’) and Preset “1” (for Principal 4’).

Added October 2010

Further additions November 2010

Click HERE to return to the jOrgan Index Page.