[PD] interpolation benchmarks (WAS: Re: better tabread4~)

Matt Barber brbrofsvl at gmail.com
Mon Jun 30 11:42:09 CEST 2008


So I thought maybe it would be a good idea to start a thread with
ideas for benchmarking the interpolation schemes.

This is what comes to mind, but I'm by no means an expert on this, so
some may be more relevant than others:

1.  Cosine wave measurements -- measure spectrum, RMS of residual
error (against which ideally-produced cosine wave?), spectrum of
error, etc.  Possible to do mathematically rather than in Pd, but Pd
would make it easier to test a lot of cases quickly.

2.  [noise~] and [vd~] (or a random table and [tabreadx~]);  measure
bandwidth of various degrees of transposition down and up.  It might
be worthwhile to test a random table with [taboscx~] and do some
spectral measurements (what happens when you make it periodic?).  Test
with pink noise, gaussian noise, sample-and-hold noise, etc.  Triangle
waves or a linearly interpolated "low-frequency" noise might be
interesting to check as well, just to see how much they distort a
straight line.

3.  Compare transposition down and up with many different vocal and
instrumental samples (ideas?).  Test residual differences between
schemes two at a time.

Please amend the list or add to it.  =o)



More information about the Pd-list mailing list