Pursuing Karplus-Strong

[Grumble]

I learned machine code on a self made "machine" built around the HD64180
I even had it sampling audio under dma, a video interface, a soundboard with a AY38910 all written in machine code.
Then I got the Junior computer around the 6502 and then a REAL computer: the Acorn Atom (also 6502 based)
Yup, same age as you

[JovianPyx]

As an addendum - There is a way to possibly speed up the low pass IIR filter which also eliminates the need for the b1 variable:

The formula can be rewritten algebraically as:

[JLS]

Update Arduino Karplus-Strong.

- better triggering
- better timers (PWM PIN 11)

Example Simple Karplus-Strong and duophonic Karplus-Automata cellular automata sequencer

---

Karplus_Strong.ino
Description:		Download (listen)
Filename:	Karplus_Strong.ino
Filesize:	1.74 KB
Downloaded:	649 Time(s)

---

---

Karplus_Automata.ino
Description:		Download (listen)
Filename:	Karplus_Automata.ino
Filesize:	2.86 KB
Downloaded:	608 Time(s)

---

[MapacheRaper]

Cool! I´ll be testing it in no time. Thanks!!

[JLS]

Floating point version Arduino Karplus-Strong.

- better sound
- better lowpass filtering
- but needs more memory and cpu resources

---

Karplus_Float.ino
Description:		Download (listen)
Filename:	Karplus_Float.ino
Filesize:	2.64 KB
Downloaded:	610 Time(s)

---

[JovianPyx]

[MapacheRaper]

That floating point version seems really cool JLS. Im diying to test it. Thanks for sharing!

@Jovian, yepp, after some weeks playing with the arduino I see you are totally right. It is more intended for gating/triggering, sequencing, etc, but for DSP it needs way more meat (even if you can do some cool tricks too, even complete oscillators)

Jovian, has you taken a peek at the ESP8266 and ESP32 chips?. It packs a lot of power and bang for the buck. Definitively I think we will see some next gen O_C and mutable type of things coming from this chips

[JovianPyx]

I had not seen the ESP chips. I don't need the wifi tho. I've had really good experience with the STM32 stuff, so I'll stick with that. Dual core is nice, but troublesome to get 2 cores cooperating usefully (parallel processing). For synth projects, a fast single core CPU is far less bother IMO. I've written multithreaded polysynths for a raspberry pi3 and that code still has issues regarding coordination of threads and memory sharing.

[MapacheRaper]

I have never done multithread, but I have meditate about it and it looks like a pain in the ass, yes. But we like it or not it´s what the future brings. It seems like 3-4GHz is the max speed that silicon does and the current state of the art chips are growing in parallel giving more and more multitreads.

So I guess somehow code will adapt to that, maybe facilitating or even completely inmaskaring in in the framework (so the framework do all the heavylifting transparently for the coder)

Regarding ESP, I see it with a lot of potential. Imagine the Ornament and Crime. Probable the more succesful open source module in existence. All people using it. A ESP version could use the wifi to connect with the smartphone/computer, so suddenly you have snapshots of the patches (infinite patches memory) plus you can control the secuencers and whatnot from the phone. And run different algoridms in every channel, something that most users ask but is quite difficult to implement in the module now.

And all this in a chip that costs 2-3USD instead of the 20-30 of the teensy.

Interesting times, for sure. I guess we will have a new wave of completely new functions in modules, no seen yet, in no time, as people start to hack with the ESP and find his treasoures

With that said, I have some SMT32 arriving in the next days too. Yumm

[remedios_dream]

@JLS
I see the topic is quite old I would like to ask for your help.

If I understand code right: assuming the sampling frequency is 22050 and our delay size is 256 we have one of 255 frequencies available, right?
Where:

freq = sample_rate / bound

where max size of bound is 255.

Is there any way to get more frequencies? The simplest one is to make delay array bigger but arduino will be out of ram then.

Maybe there's another karplus-strong implementation that do not require creating large data array?

Last one could you please explain what `accum` is?

[blue hell]

[remedios_dream]

@Blue Hell

Thank you for your answer,

You're right I totally forgot about interpolation here, also moving pointer is cool idea. I've used it several times in few wavetable kinda synthesis projects.

My main platform is stm32, unfortunately covid hits badly and those chips getting more and more expensive. I'm considering a step back, for a reasonable price.

Now I see that there's one thing which I have to learn.. How to imitate floats in weaker mcus.

There's a lot of binary shifts in code and I assume that's because we try to imitate floating point, right?

Do you recommend any resources on how to start dsp with such 8 bit micros?

[blue hell]

For non floating point .. have a look at fixed point math .. you can probably find some C macros doing the hard work for you .. although it is not really that hard, it involves some shifting ...

Only thing is with 16 bit math you need to be careful not to under- or over-flow things .. which may need some occasional rescaling (more shifts!), or using 32 bit intermediate values here and there.

And 32 bit math everywhere might be too expensive CPU wise ... and 8 bit math .. just too poor.

The shifts of course are just cheap multiplication and division (by factors of two).

[danielfiction]

[danielfiction]

Oh, i figured it out—I couldn't just change PWM Pin to 9, I had to manually wire it to 11, 3, or 9 according to the sketch.

would anyone mind telling me why that is? I usually make these changes in code without breaking functionality at all!

[quote="danielfiction"]

[herd]

necro posting, just wastin' time here,

never understood why communities were so insular.

THE reference for karplus-strong is ccrma at stanford. i kind of feel like julius o. smith III is the qi jiguang of our time, he died for your syns, notate everything, give it to the people.

even if you just stare at the pictures, julius' pages will tell you about most things you want to be aware of, to some functional level.

(one of the other prehistoric university hosted dsp pages has just disappeared, god help us when ccrma goes down)

i've only messed with k-s in a purely dsp environ. linear interpolation is kind of like a free lowpass you don't want (the algorithm is a 6dB lowpass). 2nd order hermite (musicdsp.org or at least was) at 44.1k will give you arbitrary decay lengths that are practically undamped for perhaps a minute, much better than cubic.

you want an allpass (3rd or 4th order) or more in your delay line to add "stiffness"/nice metally tones. you'll want to redact the delay line length for this.

if you get an environ where you can implement it, coupling two lines orthogonally ("vertical and horizontal vibrations") with different damping settings allows you to shape a more realistic amplitude contour. using a similar scheme with different lengths allows you to play touch harmonics.

injecting a white noise impulse is traditional but i prefer envelopes or shaped impulses (or mass-springs). you can also use a constant white noise (filtered?) for bowing effects or "fun wind instruments".

qi jiguang and julius o. smith III are your best friends ever. remember where they are.

[blue hell]

Ah, thanks for your comments herd - they might get me to do things again

[elektromeister]

I once programmed such a KS-Algo on my DSP - using this idea of multiple feed backs:

http://www.96khz.org/htm/waveguide3stage2.htm

The author states it was fine for guitars because it also creates slightly detuned third harmonics which comes more close to the real guitar.

We observe the same behavior in metal material which is bended and under torsion. Real waves differ from the mathematical ideal.

[danielfiction]

[danielfiction]

[mosc]

I love that video. A beautiful gizmo that Stan Lunetta would be proud of.

[rolfs]

[blue hell]