30 Mar
2006
30 Mar
'06
4:30 p.m.
On Mar 30, 2006, at 12:52 AM, linux-audio-dev-
request(a)music.columbia.edu wrote:
Maybe I haven't made myself clear. IMHO linear
resampling sucks
for audio. It is included in libsamplerate purely so I can
show how bad it actually is.
"Vintage" gear from the first few decades of digital audio
were largely implemented using linear resampling.
As a result, the linear aliasing characteristics have become an
"effect" that artists are actively looking for in some situations.
Ditto drop-sample resampling (Ensoniq Mirage). So I think
it makes sense to make it as fast as it can be, and present it
as an option. "Bitcrusher" plug-ins are popular for a reason :-).
This paper (sadly not available online):
D. Rossum, ``Constraint based audio interpolators,''
Proceedings of the IEEE Workshop on Applications of
Signal Processing to Audio and Acoustics, New Paltz, NY, 1993.
Shows why linear sampling can often sound "good enough"
in practice. Basically, in linear resampling some of the
high-energy aliases are masked by the signal. The paper shows
how this property isn't present for some of the resamplers that are
higher-order than linear but lower-order than sync interpolators.
So, if you can't afford to do a sync interpolator, you have to
be careful what alternative you choose, or else you may end
up with something that sounds worse than linear!
---
John Lazzaro
http://www.cs.berkeley.edu/~lazzaro
lazzaro [at] cs [dot] berkeley [dot] edu
---