The other day, I was trying to remember how to generate some simple sounds with minimal amounts of code. I remembered vaguely something called the Karplus-Strong algorithm, and it has been floating around in my head that I should look it up again. I mentioned it to Tom, and he spent 15 seconds drawing something on his whiteboard to remind me how the rudiments of it worked. I spent two or three minutes in a break typing in this small chunk of C to try it out:

[sourcecode lang=”C”]
#include <stdio.h>
#include <stdlib.h>
#include <sndfile.h>

/*
* ks.c
* a small implementation of the karplus-strong synthesis algorithm
* as described to me in 15 seconds or less by Tom Duff, and implemented
* in the two or three minutes that it took to type in.
*/

#define SAMPLE_RATE 44100
#define TONE 440

/*
* BUFSIZE is essentially the period of the sound in samples.
*/

#define BUFSIZE (SAMPLE_RATE/TONE)

double buf[BUFSIZE] ;

main()
{
int sample ;
int i, j ;
double s ;
SNDFILE * sf ;
SF_INFO sfinfo ;

sfinfo.channels = 1 ;
sfinfo.samplerate = SAMPLE_RATE ;
sfinfo.format = SF_FORMAT_WAV | SF_FORMAT_PCM_16 ;

sf = sf_open("test.wav", SFM_WRITE, &sfinfo) ;

for (i=0; i<BUFSIZE; i++)
buf[i] = 2.0 * drand48() – 1.0 ;

for (sample=i=0; sample<5*SAMPLE_RATE; sample++) {
j = i + 1 ;
if (j >= BUFSIZE) j = 0 ;
s = (buf[i] + buf[j]) / 2 ; /* simplest filter imaginable */
buf[i] = s ;
sf_write_double(sf, &s, 1) ;
i = j ;
}

sf_close(sf) ;

}
[/sourcecode]

Here is link to the single note that the above code generated. Sure, it’s not all that impressive by itself, but consider how simple the code is. It takes a small buffer of random numbers and a simple averaging operation, and generates something which isn’t a completely unpleasant plucked-string sound. I find that rather remarkable. I’ll probably use this in a little programming project that I’ve been mulling over which involves Christmas music. Stay tuned.