Subject: Re: [linux-audio-dev] So, what to do about it? (was: Still I cannotunderstand why...)
From: Lamar Owen (lamar.owen_AT_wgcr.org)
Date: Wed Dec 19 2001 - 21:03:02 EET
On Wednesday 19 December 2001 10:21 am, Juhana Sadeharju wrote:
> >Been there; done that; have no desire to do it again. Resampled 48khz to
> >44.1 khz sounds worse to my ears than the D/A->A/D converted signal.
> [ ... ]
> >IMHO, 88.2 sounds better than 96, if the result is going on Red Book CD.
> So, we need a good resampler plug-in too to avoid above problems.
Why even bother sampling at 48k in the first place if you are going to 44.1k?
Now, I can fully understand the need for oversampled (88.2 and 96) tracks for
mixdown use, but the math is pretty straightforward on showing that you are
highly likely to get problems resampling from 48k to 44.1k.
No filter is a perfect brickwall -- and you are going to have to have a tight
anit-aliasing filter ahead of the resample. The tighter the filter, the more
likely you will have problems with group delay and phasing around the cutoff
knee.
As an AM broadcaste engineer, I know a little about tight filters: the NRSC-1
filter is a bear, and is coupled with substantial preemphasis at it's knee.
The group delay is audible, if you know what to listen for. It's a good
thing most AM radios are narrowband, or that 'grunge' at 11khz would drive
listeners batty.
Although FM swish is, to me, much worse -- but my ears' high end response is
more sensitive than most.
If the filter isn't tight enough, you lose response below 22.05khz because of
it.
The distortion I have heard on resampled cuts is aliasing distortion, and is
very distinctive in nature. Sounds much like FM multipath 'swish'.
But a good resampler plugin would be a very good thing, with multiple
antialias filter settings available.
With 88.2, you simply average adjacent samples to get reasonable resampling;
96 isn't so easy. There is, of course, more than one meaning of average
here....
-- Lamar Owen WGCR Internet Radio 1 Peter 4:11
This archive was generated by hypermail 2b28 : Wed Dec 19 2001 - 20:56:36 EET