Hi!

This is a weird question, but there are some many audio pros here. I really need a piano soundboard impulse response, can anyone help me?

I've been toying with a little software piano model, and so far I've got what appears to be a respectable bass string sound. But without a soundboard response to convolve it with, I'm stuck. I do have a piano but no good recording gear.

Ole

Are you using digital waveguides for the string?

The site I was looking at that sells samples Sampletekk has says they include impulse responses for damper pedal for convolution in the samples. $49 price is not to bad.

But really, a small diaphragm made-in-china mic is not that expensive and you must already have a decent audio interface.

What are your plans for this? can you tell us what you have done so far?

RE: Sampletekk - before purchasing anything, it's probably wise to make sure that the impulse responses will be actually usable as seperate entities. I.e, it is conceivable that whilst the instruments use impulse responses of the soundboard, these impulse responses may not be able to be "exported" for use in other software. (If anyone knows either way, please advise!)

Greg.

Another thing: I think you will find that the Sampletekk impulse responses are impulse responses of the entire piano - not just the soundboard. Their purpose is to recreate sympathetic resonance, so I presume the sound is that of the entire piano with the pedal down. If you require something relating only to the soundboard itself, I will be extremely surprised if the Sampletekk IR's would be of any use.

Greg.

Thanks for the responses, although no hit yet.

ChrisA: Recording it myself is a last resort. I don't really have any experience with recording with a mic, if the quality isn't good enough, I probably can't use it.

I've currently got a model of transverse and longitudinal vibrations with hammer interaction and string coupling effects. What's lacking right now is the soundboard, dampers, sympathetic resonances (sympathetic shouldn't be too hard I think, I just haven't experimented with more than one note at the time yet) plus some of the noises occuring when you play an acoustic.

As for my plans, I want to complete the model and see if I can get a great sound out of it, then work on the performance so it can run in real-time.

Digital waveguides: No, I'm not using them. My understanding of digital waveguides is that they're extremely efficient but can't model inharmonic series directly, you need to filter them to that. But I'm a total DSP newbie, I'm not doing anything new, just leaning on some research papers. You can find some here from active researchers

http://www.acoustics.hut.fi/~hml/publications.html
http://home.mit.bme.hu/~bank/publist/index.html
http://www.diegm.uniud.it/rinaldo/workshop/Workshop/1/pres/zambon.pdf

If you're interested in how sound is generated in the acoustic piano, don't miss this one

http://www.speech.kth.se/music/5_lectures/contents.html

It's a real shame that neither Pianoteq or Roland disclose any information about their models.

Ole

FWIW, regarding Pianoteq, I just Googled "Modartt patents" and found this:
http://www.faqs.org/patents/app/20090241757

Greg.

Thanks for the very interesting links!


You do the transverse waves with three waveguides, one for each string, and the longitudinal with at least one other waveguide. At least that's the way J. O. Smith III describes how he does it. I assume you've seen his site? They even have a C++ SDK over there.

Dispersion (progressively sharp harmonic series) is really easy with waveguides, you incorporate it in the loop filter via an IIR LP or all-pass. And you can often commute the body and other resonances by using the lumped impulse response as the excitation itself, thus avoiding convolution (at least for the higher frequencies). Commutation assumes a linear model however.


It's the nature of industry to be secretive, something patents were supposed to address by providing a short-term monopoly in exchange for the revealing and protection of trade secrets. Information needs to be free IMO, and all the profit-making activity by large corporations is a huge drag on any kind of development.

Sampletekk sells "samples" not software. So there is no "export", or more accurately "export" is all you get.

The samples are available in about 5 or 6 formats and at least one of the formats is very well documented. The price is $49 for these is not much risk.

But yes, I is not clear exactly what is there. Likely you'd have to look.

Understood and agreed, however it is my understanding that the impulse response are only available for the Gigastudio 3 format, and the samples for this format are sometimes put into a single file. All I'm saying is that it would be prudent to make enquiries before actually making the purchase, to determine a) whether the impulse responses contain the desired information, and b), whether the impulse responses are usable externally, and what is required in order to do so. (e.g sample conversion software)

Greg.

I browsed through the Sampletekk libraries and did some googling. It seems the PMI Hybrid on the Sampletekk site should contain soundboard impulse responses. As you suggest, it might be an idea to contact them.

dewster: yeah, I've looked at Julius Smith's work, that's actually where I started. The Finnish research is based on digital waveguides I think. It's just not the only way to do it, for real-time purposes it might have been 15 years ago but processors are faster these days. I also looked at the SDK, but I had some beefs with it and the examples had a tendency to segfault on me.

Greg: thanks for the link to the patent application, hadn't come across that before. I'm reading it right now.

Found the patent application on freepatentsonline.com. At least half of it is pure crap, I certainly hope it's not awarded. From what I could gather from it, it seems like Pianoteq is using simple modal synthesis with some kind of prerendered soundboard model. It'll be interesting how the model I'm working on stack up against it, once I get the soundboard response.

I waded through that swamp too, though you no doubt understand it better than I do. To award a patent my criterion would be full disclosure of exactly what they are doing, not a couple of equations and a bunch of hand waving. I guess business doesn't want that though (but they do enjoy the protection a patent gives them). I think patent lawyers get paid by the word.


Several years ago I ran across some sound clips on the web that demonstrated how a narrow pulse waveform could sound uncannily like a violin via filtering - it was quite fascinating and I wish I could find it again.

This is a very interesting thread to me, please do post any progress that you make!