18 May
2010
18 May
'10
11:46 a.m.
On Tue, May 18, 2010 at 14:11, Liam Proven <lproven at gmail.com> wrote:
Actually, my guess it it wouldn't do nearly as well as you think,
without a little (in fact, trivial) preprocessing. Basically,
digitizing the tape would mostly digitize the tape hiss, which can be
viewed as white gaussian noise. This doesn't compress with an entropy
(lossless) coder very well (at all). Assuming the hiss stays within
maybe the lowest 16 quantization levels, you'll still need 4
bits/sample, and no algorithm will get around that. (I'm assuming 16
bits/sample common to modern soundcards)
Solution: use fewer levels, 8bits/sample will probably do (but don't
use uLaw/ALaw). Better yet, use a noise gate clamping all samples
below some threshold to 0. Heck, with some formats, 2 bits/sample
with a well-chosen quantizer would probably work. Choose a
appropriate sampling rate and you can do better than any
audio-oriented compression format, lossy or lossless.
Joe.
--
Joachim Thiemann :: http://www.tsp.ece.mcgill.ca/~jthiem
If you want to
store the data uncompressed, it'd be worth thinking
about the sampling rate a bit: ?my guess is that standard phone rate
of 8000 Hz would be sufficient. ?But it depends on the modulation
scheme the Apple uses.
FLAC <> uncompressed. It's just not /lossy/ compression. It's still
highly compressed and on something like a datasette, it should achieve
fairly stellar compression rates, I'd think.