Does ChucK Play the Piano?

Inventor

For my relaxation series and maybe even for the rock, I'd like to add some piano. What's the best instrument / ugen for this?

kijjaz · Posted: Thu Mar 13, 2008 9:20 pm Post subject:

Rhodey and Wurly in the StkInstruments sound quite real.
But Wooden Piano..? hmm..

I'm thinking about using some digital waveguide synthesis for the piano.

Inventor · Posted: Thu Mar 13, 2008 9:42 pm Post subject:

You know digital waveguide synthesis, kijjaz? That's way over my head. Your ChucK powers are strong indeed. I'll check out Wurly and Rhodey for Guitar Lab, thanks.

kijjaz · Posted: Fri Mar 14, 2008 1:02 am Post subject:

Hmm.. i'm not that strong at doing something deep heheh -_-"..
(but i am crazy)

I'm going to work on some general digital waveguide models.
Maybe I'll start with a class that manipulate an array of floats that'd be used for the models i'll come up.

Frostburn · Posted: Fri Mar 14, 2008 5:50 am Post subject:

http://ccrma.stanford.edu/~jos/wg.html has research papers on Digital Waveguide synthesis, including modeling the piano. Most of the models use a sampled or otherwise tabulated hammer impulses and it's not easy to make a realistic piano without them.

However the bulk of "synthesizer" pianos are just samplers and some of the soundfonts I've found aren't half bad. Maybe we/you/I could turn a free soundfont of choice to a class with SndBuf's in chuck.

Inventor · Posted: Fri Mar 14, 2008 8:15 am Post subject:

Wow, that seems to be a very well written set of documentation for digital waveguide synthesis. It is quite similar to the waveguide simulations that I did in college. Thanks for the reference, Frostburn.

Hmmm, so we must decide between samples and waveguides? Well, we are up against compute limits unless I can optimize the program somehow, so samples look nice. Then again, a wavegiude model could be quite efficient. I wonder about the situation of new notes striking while old notes are still resonating. The sequencer in Guitar Lab strikes a note every so many times per second, whether the previous note is finished or not.

To accommodate that might take an entire piano of waveguides, which would be too inefficient, then again, is SndBuff capable of dealing with overlapping samples?

I like the idea of keeping the project to a single file with no subdirectories for sound files, but if a zipped up directory is what we need, then I'll have to figure out how to make tar and gzip wrap them up into a single file for distribution (that's no problem).

I don't know, both solutions look equally menacing to me! haha! Would either one handle the overlapping notes issue better than the other? Perhaps the answer is to have both models in the program.

Frostburn · Posted: Fri Mar 14, 2008 8:30 am Post subject:

Inventor · Posted: Fri Mar 14, 2008 9:04 am Post subject:

What the program does is it has a button for each instrument. When an instrument's button is toggled on, the instrument gets ChucKed up, and when it is toggled off, the instrument gets unChucKed. The SndBuf.read could be called at the button press moment, so no problem there. 256 of anything sounds like a lot, but maybe it's manageable, and we may as well just do mono since the rest of guitar lab is mono.

I guess we should try both models and see what happens, eh?

Frostburn · Posted: Fri Mar 14, 2008 12:19 pm Post subject:

Another option besides waveguides or samples is additive synthesis. That is controlling the spectrum in time with float functions. It's not as fast but it can still be done in real time with IFFT.

I'd start with the sampler but if I make it I will make it as an Instrument in my library. (I don't have a Mac so I can't run MAUI and coding for something I cannot debug would be a bit strange.)

WG and additive synthesis take a lot of skill if your goal is the piano. The instrument is only 300 years old and it is the product of great cratmanship and fine tuning. It's timbre is not easy to replicate but trying is always fun. :)

Inventor · Posted: Fri Mar 14, 2008 12:43 pm Post subject:

Oh, your instrument library sounds interesting, perhaps there are some other instruments in there that I could add to Guitar Lab. Maybe you've already got some goodies in there that are suitable for re-use.

The additive synthesis sounds neat, but how would we know what time-varying functions to apply to the IFFT to get a piano?

Frostburn · Posted: Fri Mar 14, 2008 12:56 pm Post subject:

radarsat1 · Joined: Mar 14, 2008 Posts: 14 Location: Montreal

hi,

maybe someone would be willing to port this to a ChucK Ugen. I wish I had time to do it myself but I don't. It's STK, so it should be fairly straightforward to port it.

http://www.music.mcgill.ca/~sinclair/content/stk_piano

cheers.

Inventor · Posted: Fri Mar 14, 2008 1:27 pm Post subject:

I skimmed through the pdf on that stk_piano model. I had no idea I was suggesting such a complex and difficult thing to do. Perhaps we can just borrow the hammer model from it, which is an impulse followed by low pass filters. Then maybe borrow some pieces of the waveguide stuff, and make a "piano-like" instrument.

Rather than painstakingly model all the characteristics of a real piano, we can just use a simple model to approximate it. This keeps the cpu usage low and the amount of effort low as well, while probably producing fairly good sound.

radarsat1 · Joined: Mar 14, 2008 Posts: 14 Location: Montreal

well, my point was more that it's already done. but sure, feel free to pick it apart too if that's helpful.

Frostburn · Posted: Sat Mar 22, 2008 5:25 am Post subject:

I just tried the additive synthesis approach.
So far too thin.
A good grand really needs that internal reverbation to give all that space and brightness to the sound.

I skimmed through the pdf for Stk piano. Looking good. I still haven't tried it but if it sounds ok it would be worth the port.
Anyone with some stk experience and too much free time? Smile

Kassen · Posted: Sat Mar 22, 2008 7:07 am Post subject:

I'd go for parallel band-pass filters for the piano's internal reverberation, I think, but I have a soft-spot for modal synthesis anyway.

The internal reverberation makes a huge difference indeed, just for kicks add this to your favorite cheap toy piano sound;
http://www.pspaudioware.com/plugins/piano.html

radarsat1 · Joined: Mar 14, 2008 Posts: 14 Location: Montreal

i just tested stkpiano since i haven't actually tried it in a while.
At the default settings there was some serious distortion in the output.
However, i tried with "./pianotest voices=1 buffersize=256" and it sounded fine, so I recommend those settings if you want to give it a try. I guess one bad thing about it is that it's not exactly optimally efficient. There are a lot of extra filters and things in there to make it sound marginally better that could probably be sacrificed for speed.

Who knows maybe in the summer I'll have time to do a port to ChucK, it would be great.. but I can't promise.

By the way if you haven't read the document... the gist of it is that it uses waveguides for notes lower than 88, and modal synthesis for the higher notes. It uses look-up tables for the non-linear velocity mapping. And there's a bunch of all-pass filters to model stiffness... it can be made to sound like a harpsichord if you turn this up.