[flac-dev] Next generation WebM and FLAC

Martin Leese martin.leese at stanfordalumni.org
Wed Jun 20 12:20:55 PDT 2012

James Haigh <james.r.haigh at gmail.com>

> Note that 50% is silly, lossless compression is asymptotic. I 2nd Martin,
> once you have high-density entropy, there's little more 'air' to
> squeeze-out. 10-20% would be worth it if it helps adoption, although it's
> worth studying how close we already are to the asymptote of entropy. How
> much would be saved? How long would it take? Is it really worth it?

There is more to compression than simple
compression.  Are there any other features
that are missing from FLAC?

I am thinking, here, about JPEG 2000.  The
increase in compression over JPEG (1992) is
quite modest.  What sells the format is other
features such as combining lossy and lossless
in a single standard (by flipping the wavelet),
Region of Interest to store different parts of the
same picture at different qualities, the ability to
send a lossy image first and then progressively
refine it until it is lossless, etc.

Direct audio analogies of these features don't
seem appealing, but is there something else?
Dennis has suggested 32-bit float audio.  As
this is becoming popular in production,
accommodating it would aid convenience.

> Can wavelets be applied to lossless sound compression? I've seen that
> they're used for JPEG 2000, Dirac, and was planned for Tarkin, all of which
> image compression standards. But what about audio?

Good question.  Provided the expectation is an
improvement of only 10% to 20% then this
would be a useful research exercise.

Martin J Leese
E-mail: martin.leese  stanfordalumni.org
Web: http://members.tripod.com/martin_leese/

More information about the flac-dev mailing list