Latest Updates on Progress...
I am making great progress at the moment, with another Beta version very close to being completed. On the list to complete is to validate the Swell and wind models, fix a few issues with a couple of tremulant ranks and finish off the custom organ rank definition files. In response to some questions on the Hauptwerk forum regarding the piston configuration on the console, see below a rather rough and ready image of the real console showing the piston layout. Note that there are 8 general pistons with 128 levels of memory (1-4 on the Left hand side of the solo piston rail, and 5-8 on the left hand side of the Swell piston rail, and 8 general toe pistons which also double up as the Swell divisionals at the flick of a switch), 8 divisional pistons with 8 levels of memory, and various sequencer frame + and - pistons on the Swell and Great. There are also reversible thumb pistons for coupling each manual to the pedals, Swell to Great and Swell to Choir, and reversible Foot pistons for both the 32' pedal ranks, Swell to Pedal and Great to Pedal. I hope that answers the question, but please feel free to contact me if you have any other questions!
I have completed the graphics and programming of the registration sequencer (It now works like the real thing with the LED display showing the frame and bank numbers and all the buttons (bank + / -, locks etc.) working as they should. The samples are complete and I am currently fine tuning the markers and timings for start and release. This is incredibly time consuming, as it involves going through every single sample one by one and ensuring they are correct - I estimate a couple more weeks to complete this. The Swell tremulant (which was not functional during the sampling sessions) has been re-created using recordings taken subsequent to re-pitching using the hauptwerk tremulant model. There is still some work to do to fine tune the results here, but the hard work is complete. The Choir and solo tremulants are in almost all cases using direct tremulant effected samples (which in theory could be used to create Hauptwerk sample models if required) The logic of this is that there is more realism as the tremulant is heard correctly in the accoustic (which is a problem with the modelled tremulant), although this is at the expense of synchronisation of the tremulant between notes - I would be interested to hear peoples comments on the relative merits and realism tradeoffs of each method! The next job will be to fine tune the swell and wind models (which are only set up roughly so far) using data derived from reference recordings. I then need to get the whole lot off to my Beta tester volunteers to see what they make of it! At which point I need to start building some of the website functionality to host downloads and manage payments etc.
The time and attention to detail involved in a sampling project like this is enormous. Having developed software to manage many of the more laborious tasks (With the robot recording the notes and placing the markers, and the software doing the bulk of the denoising and preparing the samples, I can now create a rank with very little human intervention (save for a manual pass to edit and clean up after the denoising processes, and a manual check on marker placement). Without these aids, I fear I would simply not have had time to devote to this marvellous project!
A note on generating ambient samples: I have spent a great deal of time optimising the process of removing noise from ambient samples. The problem as I see it is that the reverb tail gets gradually quieter and eventually falls below the noise floor; In a large ambient building, this happens relatively slowly, and as the reverb tail gets quieter, the noise is relatively louder in comparison, to the point that it is easily heard by the listener. Because it is necessary to remove almost all the noise at this moment, particularly as it is not masked by something louder (a note from a pipe), one has to be quite brutal with the noise reduction to prevent noise being heard. There is also the issue of attempting to re-create the missing piece of reverb which is not possible to record due to it being masked by the noise floor in the building. My solution is twofold: Using progressive noise reduction, whereby the amount of noise reduction applied varies - less where we want more detail (during the pipe speech for example to preserve those lovely edge tones etc.), and more where required to prevent noise being heard (at the end of the reverb tails). The missing reverb tail which was lost under the noise floor during recording can then be reconstructed using close up Stereo recordings of the pipes and matching stereo impulse responses which are recorded from a number of locations within the pipe chambers. The resulting sample reverb tails are then produced using the real building acoustic just prior to the point where they are lost in the background noise, and then reconstructed using impulse responses to generate the missing audio data to the end of the tail, resulting in a highly accurate and noise free ambience. Obviously attempting to do this manually for each sample would not be practical, but fortunately, with a bit of programming, software has come to my aid!