Sergey Batov wrote:

> Here are two examples of Karplus Strong Algorithm (from List).
> 
> It seems that the pitches of notes  are good enough in first example ,
> but not in second example. Why?
> In the beginning I thought that this is because of LPF y[n]=(x[n]+x[n-1])/2
> in the second example, but using the 'tone' brings worst results.
> 
> May be the pair "delayr - delayw" works more exactly than "delay"?

Try using deltapi in conjunction with a delayr/delayw pair.  delay will
round the pitch off to the nearest integer sample number, which is bad
for pitch accuracy at higher pitches.  deltapi will interpolate between
samples to get a more exact pitch, although the slight amount of delay
of the LPF in the feedback loop will probably add a slight bit of pitch
change.  You should also note that the linear interpolation in deltapi
will act very close to the LPF you describe in your example, so the tone
and sustain of pitches will probably vary in some unpredictable ways.