[josephernest]

Oh something in my niche ! I produce French pipe organ sample sets for a living : https://www.jeuxdorgues.com

It's an extroardinary journey to record an organ, process the thousands of WAV files and design a virtual organ model.

[isoprophlex]

I know what a WAV file is, and I've even made some electronic music myself.

Yet my brain errored out and for a while I thought you made samples of pipe organs: little mini-organs to take home with you, to try out before you buy the big one.

"Hello! Can I interest you in trying one of our sample organs?"

[wlonkly]

You've reinvented the portative!

[vunderba]

Very cool! I'd be curious what you think of Modartt's recent attempts at physical models of organs and how they compare to more traditional sampler approaches.

https://www.modartt.com/organteq_physical_modeling

[smj-edison]

Not OP, but I've worked on programming my own organ software before. I'd say physical modelling really cool tech, but the question is whether it makes a significant auditory difference.

The thing is organs are a lot easier to sample compared to something like a violin. I'm oversimplifying, but it's mainly just note on, note off, vs lots of articulations where physical modelling is more beneficial. (Yes, there's wind sag, and wind delay, and regulators, but most organs specifically have things to avoid those artifacts so they'd only show up on more niche organs imo.) I've had great success simulating tremulants by just using FM demodulation to reconstruct the pitch and volume effects from tremmed samples[1]. Release samples are also difficult to match with the current phase, but I was also able to mitigate that with a single bin DFT + crossfaded.

Another issue with physical modelling is it's decently CPU intensive, which is tricky when you have 700 simultaneous notes on bigger organs. So, it's definitely cool, but the question is whether it's significantly better than current sample-based technology. It could potentially reproduce some of the more strange interactions, but those interactions aren't necessarily wanted in the first place.

EDIT: one thing that is nice about physical modelling is it's a lot easier to voice (modify) a pipe to the sound one wants. I think with some special filtering (comb filter for even harmonic attenuation, shelf for augmenting the harmonic series trajectory) voicing could also be satisfactory with traditional sampling (hauptwerk does some of this, but I think I could make it even more flexible).

[1] https://github.com/smj-edison/sample-analysis

[tgv]

Not the one you asked, but I have both Hauptwerk and Organteq, and the former beats the latter on everything expect cost, memory usage, start-up time. Organteq sounds artificial, and its reverb doesn't make it any better. Ok, you can tweak a lot, but the ensemble sound is unconvincing, and the registration choices are too limited.

[tptacek]

What! For a living? Say more!

[HPsquared]

People do that with pianos and other instruments too. Some of these sample packs are very expensive. It is a small market though, I suppose.

[pontusrehula]

I'm interested in your choice to focus on French organs. Do you specifically aim to focus on French organs and not those of German type? Or is it more like a matter of convenience due to geography?

[wglb]

Let's hear more!

[josephernest]

You can hear many MP3 examples here: https://www.jeuxdorgues.com/jeux-d-orgues-4-lachaisedieu/

All are played with the virtual organ, what do you think?

I'll do a post soon with more stories about this project itself (that I started long ago)!

[mock-possum]

What?? That’s such a specific thing to do!