[linux-audio-user] Re: Resampling Audio Libraries & Sinc

i happened to have a short glimpse on your posting,
david. i don't want to insult anyone, but there is a
bit more to evaluating accuracy than comparing sample
values at fixed locations ;--)))))))

consider a few of these things :

a) non-causal algorithms introduce a natural delay.
for windowed sync interpolators this is half of the
filter kernel. ok, the algorithm could skip those
samples of course (that's what fscape is doing by the
way, if i remember correct).

b) the algorithm could introduce a gain factor
different from 1.0, this is in fact very likely,
consider resampling 44.1 khz to 22.05 khz, a 0 dB
headroom file would be distorted if there is no gain
compensation.

c) the crucial point about windowed sync is that it's
naturally also a lowpass filter (for upsampling which
you were doing in your example). so, those filters are
windowed sync filters, they are linear phase and show
an effect called gibbs effect which means "edges" show
pre and post ringing phenomena. this is not
necessarily "bad" and with not extremely long kernels
it's not audible, too. in fact, when you run a
software synthesizer you will most likely want band
limited oscillators, i.e. those with gibbs effect.
now, no realizable filter is infinitely steep, so
you'll have a passband etc.

summa summarum it's very likely that for an arbitrary
input sound, resampling with an integer factor will
*not* produce equal sample values with the factor
spacing in the output.

without blowing up this stupid discussion, i suggest
to do two things :

a) *listen* to the results for an aural comparison.
this is in fact the most important point, because for
different ears different algorithms are better or
worse. i, for example, use bandlimited resampling in
most cases and in real resamping cases (e.g. 48 kHz ->
44.1 kHz), but sometimes i prefer the additional
colouring and brightening (= distortion) of say cubic
interpolations or even linear interpolations which
produce the hardware-sampler-kind-of-timbre.
b) read a book about audio analysis, and perform some
more appropriate tests, for example calculate the
harmonic distortion (try your example with a sine
tone!).

best, -sciss-

        
                
___________________________________________________________
Gesendet von Yahoo! Mail - Jetzt mit 1GB Speicher kostenlos - Hier anmelden: http://mail.yahoo.de
Received on Thu Aug 25 00:15:14 2005