[vorbis] Partial Format converters?

[Keith Wright]

> From: "R.J.J.H. van Son" <Rob.van.Son at hum.uva.nl>
> 
> > On Mon, 2002-08-19 at 11:55, Mark Hetherington wrote:
>
> > It is not possible to directly convert from any compressed format
> > to a different compressed format without decoding->encoding. It is
> > a logical impossibility.  The data is of a completely different
> > type.

Surely that is overstated.  It is possible to imagine pairs of formats
which could be interconverted without full decoding.  But this is a
quibble---it's not logically impossible, but nobody can do it with the
formats now in actual use.

This may be too obvious, but re-encoding need not involve creating a
huge decoded file and then encoding that.  The decoder can pipe data
to the encoder with a fairly small buffer in fast RAM.

> As I suspect that some of the problems have to do with the
> introduction of quantization noise (both in amplitude and
> frequency), there might still be something to win if we could do a
> half-way decoding/encoding.

If you must re-encode, don't decode to a .wav file at the sampling
rate and resolution needed for the sound card and speaker you
happen to have now.  Decode to as many bits at as great a sampling
rate as you can stand, then re-encode that to a sampling rate
that is an divisor of the rate of the intermediate representation.
(Or rather pick the intermediate sampling rate to be the greatest
common multiple (:-) of the source and target rates.)

> But if I have the idea that there is something to win over just
> doing/refraining from repeated compression, I can make a point for
> (or against) building translators to people willing to hire someone
> to actually code it.

Under no circumstances do repeated compression.  Save (and back up)
the originals, whatever form they may be.  If you must compress to
distribute the files or to keep them handy on-line, compress from the
originals.  When you switch to a new format in the future, go back to
the originals.

> I mean, all compression starts (and decompression ends) with a
> spectral (band-filter) representation of the sound of some kind.

This is false or trivial, depending on how loosly you interpret the
phrase "of some kind".  Some speech compressors code motions of
a model of the human vocal tract.

> I know that this is not easy stuff, but I have seen some very
> complex things accomplished.  If it would be worthwhile (which is
> rather uncertain), we might get corpus groups to hire some people
> who DO know about these things.  ... A "spare" programmer is rather
> cheap in these circles.

Unfortunately there is more to it than programming.  The current
copyright and patent laws are slanted toward the interests of large
corporations that want to make big money by "owning" other peoples
songs and data formats.  The needs of historians, scientists, and
others who want to store and exchange data in a form that can be used
in the future without asking permission or begging for support are not
given much weight.  You have not only to find programmers who know how
to write the program, but a team of lawyers to defend them.  This is
the problem Vorbis may solve.  It is not a good intermediate format
for re-encoding between lossy compression codecs, but neither is
anything else.

God grant that in the future we do not find that the only recordings
of spoken Navaho are in a format that was proprietary to a company
that went out of business and is no longer decodable.  I've heard
that certain census records from the fifties have already met that
fate.


-- 
     -- Keith Wright  <kwright at free-comp-shop.com>

Programmer in Chief, Free Computer Shop <http://www.free-comp-shop.com>
         ---  Food, Shelter, Source code.  ---
--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.