Understood Michael.  The underlying mathematical principles of digital
manipulation are the same for all data stored as 1s and 0s, but I'm
guessing that the same processing techniques are used with slightly
different objectives in mind depending on the application.  For
instance, in digital audio, noise shaping is used to apply dither in
such away that the random noise is applied less in the most sensitive
frequencies of human hearing and more in the very low and the high
frequencies where sounds need to be louder to be audible.

Also, I think that digital audio is a lot more sensitive to minor
errors and that you can't get away with the same approximations that
you can in the video/imaging world.  In other words, it is possible to
*hear* non-linearities and correlated noise which you wouldn't be able
to *see* in an image (or which wouldn't be so objectionable).

I got a little confused earlier, but I'm coming round to the fact that
stepping up in word size from (say) 16 to 24 bits is a trivial
operation if you just insert zeroes.  On the other hand, resolution
enhancement, as performed by the Meridian 518 and lots of other
high-end audio devices, is something "extra".  As I posted earlier,
'this white paper from Wadia'
(http://www.wadia.com/technology/technicalpapers/Resolution_Enhancement.pdf)
gives a very good description of resolution enhancement and shows why
one might want to step up from 16 to 24 bits -with the addition of
dither-.

As far as up- or re-sampling is concerned, the samples *are* modified
in a non-trivial way and I am certain that applying dither is a
*mandatory* part of the process.

I urge you to read read pages 37-40 of the 'Meridian 518 handbook'
(http://www.meridian-audio.com/download/Handbooks/500_Series/518user.pdf)
(if not the whole thing).  It's not an easy read, but there is so much
information in there about digital audio processing from one of the
true pioneers and experts in the industry.

BTW, I will be experimenting with the following sox command line later
today.  Because ringing is a product of steep filters, I've decided to
try a more relaxed filter with a bandwidth of 90.7% (which corresponds
to an (IMO) acceptable upper filter frequency of 20kHz).  I'm also, for
the time being going to allow aliasing beyond the upper cut-off, but I
may turn off that option if the bandwidth relaxation reduces ringing
significantly.  I'm also using the sox equivalent of Meridian's "high
pass" TPDF dither.

I'm not using any volume modification at present, and I'm reluctant to
do so unless I find that it is absolutely necessary.


Code:
--------------------
    flc flc squeezebox 00:04:20:10:14:eb
  # FT:{START=--skip=%t}U:{END=--until=%v}
  [flac] -dc --totally-silent $START$ $END$ -- $FILE$ | [sox] -q -t wav - -t 
flac -C0 -b24 - rate -v -M -a -b 90.7 88200 dither -S
--------------------


-- 
VirusKiller
------------------------------------------------------------------------
VirusKiller's Profile: http://forums.slimdevices.com/member.php?userid=19799
View this thread: http://forums.slimdevices.com/showthread.php?t=69454

_______________________________________________
ripping mailing list
[email protected]
http://lists.slimdevices.com/mailman/listinfo/ripping

Reply via email to