Hi All,

I've just released a free app for iOS called Samvada that uses Pure Data, through libpd and pd-for-ios. It's meant to simulate sympathetic strings, like on a sitar, for practicing and performing Indian music. It's also fun for other uses, and the strings make a nice resonance turning background sounds into tones. You can watch the demo and download it here:

Samvada for iOS

It's basically a bank of Karplus-Strong resonators. The UI handles the tuning aspects, which are aimed at Indian musicians using a drone and the Sa Re Ga Ma system of note naming/notation.

I'd welcome any feedback or questions about the process of making an app like this, etc.

Enjoy!

Very good and interesting app! It's almost the same thing, I wanted make(and after I started I found this). Microtuned drones is absolutely magical thing, I even thinking to make a real monochord to make such drones. Also, very good visual design, it's outstanding for a non-commercial app.
Just asking, would you share your Karplus-Strong resonator module? I noticed resonators doesn't have jivari. I'll maybe try to add it, but I don't know how resonator could be made by this method.

upd:

Updated: May 23, 2013
Improved 'sitar-like' jawari tone

LOL, that's funny. And now I even more want to look at your synthesis module. Pretty please! :3

Hi hodok,

I'm glad you like Samvada - thanks.

Unfortunately it's a bit hard to send you the core string synthesis patch as it contains a custom external (2nd order allpass filter) and currently relies on messages from Objective-C for much of its functionality. However, I can describe what it does. There are two tricks used that add a bit of jawari tone to the basic Karplus-Strong waveguide:

1. Feeding a (small!) amount of the output back into the delay length of the waveguide. This simulates the sitar's flat bridge, slightly changing the length of the string with each vibration. Watch out though, more than even a tiny amount will create the most horrible digital feedback sound ever.

2. String inharmonicity. Adding an allpass filter in the Karplus-Strong loop (I used a second order one) will do this. For some reason the combination of this with point 1 above results in something like the jawari sound (though it's not perfect).

You might want to check http://www.talaash.at/sitar/ which has a great sitar model. I had to cut a few corners in my version: cramming 13 oscillators into an iOS app is pushing things in terms of performance.

Let me know (here or on email) if you have any other questions.

i must have been in la la land when this was first announced. installing now.

iotic
Thanks for your valuable advices. For a couple of days I tried to recreate that algorithm by Sadjad Siddiq, but I found what it's too difficult for me for now, and I decided to try that little trick with Karplus-Strong algorithm, but the next problem has appeared: now I'm using Max/MSP(because of Max for Live), and I'm using snapshot~ to convert output signal to float number and feed it to number input of tapout as the delay length. Snapshot~ is set to the time resolution of 1 sample, and it freezes the application, but I've been advised to use gen~ to avoid that somehow, so I think I'll manage.

I don't know if you fixed this hodok. In order to get pitches high enough, i had to use a block~ 16 just around the core K-S synthesis unit.

This is a little late, but can I ask how you're modulating the delay lengths? Just adding some feedback from a [vd~] output to its own delay time input would result in a dsp loop, and using [s~] and [r~] or whatever would delay the feedback by one block, which probably isn't the effect that you're looking for. The easiest way I could think to do it would be to have two delay reads that are the same (nominal) length, with one fixed one modulating the other. I think this would be close, but not exactly accurate.

Hi acreil. Since I had needed to keep the patch as fast as possible to run 13 resonators on fairly low-performance devices, I cut as many corners as possible. So yes, I used [s~] and [r~] objects to feed the one [vd~] back to its [delwrite~]. No it's not ideal, and you have to keep the "jawari" amount pretty low to avoid some really nasty feedback, but that's the best I could get away with under the performance limitations. I know others have used a model as in your suggestion - two separate delay lines - and this seems a good idea if you have the CPU available.

That makes sense. A block size of 1 would work best, but it would be too expensive.

What struck me as an interesting possibility with the second delay tap case is that it could be set to some different length. Of course this has no basis in reality, but it could be an interesting way to change the phase of the delay modulation, or possibly emphasize certain harmonics.

I'm also imagining that you could crudely model the bridge shape by inserting a nonlinearity between the delay read and the [vd~] delay time input, and maybe also insert a [lop~] or similar. I think this might help to avoid unbounded oscillations.

In general it seems that you can do lots of abstract and more or less arbitrary variations to basic physical models with the result not necessarily sounding obviously synthetic. I'm consistently surprised at how easy it is to make something sound plausibly acoustic, even if it doesn't necessarily suggest a specific instrument.

Hey.

• different length delay modulating the string length. Yeah, I guess what you'll get there is like FM sidebands, except with both carrier and modulator resonating according to the input.

• nonlinearity in the return. Yes, I used that idea in Samvada - a cubic soft clip, I think.

• "acousticness" - absolutely. "Physical modelling" sounds impressive but it's really very simple elements. It's amazing how organic the results can be.

I'd read previously that certain percussion instruments (cymbals and asian instruments like tamtams) exhibit nonlinear behavior that is audible as a pitch glide or increasing brightness over time. This is due to thin metal transmitting sound faster as it deflects. This can be demonstrated by exciting an ordinary cymbal with a sinusoidal signal. As input level increases, additional harmonics and subharmonics are added, and eventually this breaks into chaotic oscillation.

What this suggests to me is that maybe these effects could be crudely simulated by adding delay self-modulation (with the absolute value of the modulating signal) into a feedback delay network.

And I guess this could also incorporated into other configurations. I often use small FDNs with 3x3 rotation matrices to obtain various metallic and resonant effects. This makes me wonder what kinds of effects I could get this way, either for "instruments" with impulsive or noise-based input or resonators processing other sounds.

That's an interesting idea acreil. I don't know so much about pitch glides in cymbals, gongs etc, but it occurs to me that something like that might be an interesting way to model another kind of Indian jawari - the jawari on a tabla head. As I understand it, there's a thread that goes round the part of the rim of the tabla that acts in a similar way to the jawari on a sitar or tambura. I don't know if much work has gone into modelling tabla, but it would certainly be an interesting project.

Still looking for realistic sitar physical model, any suggestions?

Yop,

simply couldn't help it.
Thank you Iotic, you inspired me a lot (as all forum contibutors here).

So I tried to make a Samvada-inspired patch , and the topic is http://forum.pdpatchrepo.info/topic/8269/sympathetic-strings-resonator-inspired-by-samvada-app .

So fun when you learn things and get immediately rewarded with a simple yet functional patch

Suggestions welcome !

Nau