Kasper

I am building a simple Karplus-strong thing in max6. So far so good, it works, BUT when I play "higher" notes it goes out of pitch (really badly). I understood it is a problem with the I/O an d signal vector size and how it affects the delay line which creates the sound. So I decided to put the whole delay line (tapin~ / tapout~) into a poly and upsamle it or something (so it is a question about the upsamplig of the poly~)

when i create a [poly~ mypatch vs 2] (the delay being "mypatch") it seems to work - at least it is WAY better

_is it "better" to reduce the signal vector size (down to 2 ?) or to increase the sampling Rate (up 2 or up 4 ???)

I am building a simple Karplus-strong thing in max6. So far so good, it works, BUT when I play "higher" notes it goes out of pitch (really badly). I understood it is a problem with the I/O an d signal vector size and how it affects the delay line which creates the sound. So I decided to put the whole delay line (tapin~ / tapout~) into a poly and upsamle it or something (so it is a question about the upsamplig of the  poly~)

usually  my set up is
I/O vector size 512
signal vector size 258
SR 44100

questions :
_is it "better" to reduce the signal vector size (down to 2 ?) or to increase the sampling Rate (up 2 or up 4 ???)

up-sampling-in-poly-or-vs-reduce

the issue is vector size and sampling rate. but mainly vector size. so setting your poly~ at vs 2 will enable a far better karplus strong algo. however, it will also increase your cpu load so dramatically that it is almost pointless. it is very difficult (almost impossible) in environments with fixed vectors in signal processing (e.g. MSP, Pd, Supercollider, Csound, Reaktor, etc) to make good quality full range karplus strong algos. some environments have workarounds with varying degrees of success. in MSP the workaround is to write your own externals in C using the SDK.

luckily, in Max6, there is not a workaround but a solution. it is called gen~. it works. there are even karplus strong and waveguide examples for you in the gen~ examples folder that comes with your distro. it is very easy - the patching is almost identical to how you would achieve it in MSP-land.

gen~ operates at 1-sample feedback rate (rather than a variable 'block' of samples in a vector) and also offers various means of interpolation etc. so kp algos are just one of the many things possible.

the issue is vector size and sampling rate. but mainly vector size. so setting your poly~ at vs 2 will enable a far better karplus strong algo. however, it will also increase your cpu load so dramatically that it is almost pointless. it is very difficult (almost impossible) in environments with fixed vectors in signal processing (e.g. MSP, Pd, Supercollider, Csound, Reaktor, etc) to make good quality full range karplus strong algos. some environments  have workarounds with varying degrees of success. in MSP the workaround is to write your own externals in C using the SDK.

even if the "vs 2" message gives me good results (an does not seem to be THAT heavy on the cpu - it is below 2%, with just one Karplus) it also gives me a good reason/starting point to have a look at gen~

actually I was doing this karplus strong thing just to understand how it's done and why it was not in tune (and then, who cares if a simili-guitar-string-feedback is in tune ?), but finding a way to start using Gen~ is a great thing

yes, if i hear "karplus strong" i also say it is both.

the smaller vector is required to be able to make shorter delay times, and the higher samplingrate is a matter of the soundquality required. if you want vectors of 2, 4 or 8, well, depends on the application.

for building a a guitar like instrument where 91 is the highest note a vectorsize of 8 might be cool.

yes, if i hear "karplus strong" i also say it is both.
the smaller vector is required to be able to make shorter delay times, and the higher samplingrate is a matter of the soundquality required. if you want vectors of 2, 4 or 8, well, depends on the application.
for building a a guitar like instrument where 91 is the highest note a vectorsize of 8 might be cool.


up-sampling in Poly - or VS reduce ?