1 Aug
2007
1 Aug
'07
4:09 p.m.
On Wed, 2007-08-01 at 10:45 -0700, Gregory Alan Hildstrom wrote:
I got 32-bit true lossless j2kaudio transcoding
working today. I compared it against wavpack and
wavpack seems far superior in terms of speed and compression for 32-bit floating point
data. I
created the test files with audacity, which only uses float values from +1 to -1, so I am
not sure
if it actually uses the full 32-bits, but it is still good enough for a meaningful test.
Here are the compressed size % for the worst case that I tested:
32bWN.wav 44.1kHz/32b/10ch
60sec 10 channels of white noise
105840152B
j2kaudio 91.06
wavpack 82.32
tar czf (gzip) 92.29
tar cjf (bz2) 93.24
The full test tables are here:
http://geocities.com/hildstrom/projects/j2kaudio/
Are there any wav/audio programs out there that use the full float range of 3.4e+/-38? It
seems to
me that this larger range would use more of the 32 bits.
all of them :)
the idea is that 0 ... +1 maps to the nominal range of the D/A converter
(-inf dB ... 0dbFS), leaving the rest of the range to span larger and
smaller values without clipping.
as fons explained recently, you could agree on a different convention
for this, and say that 0 ... 2^23-1 mapped to the physical range, but
you don't gain anything in so doing.
--p