[Vorbis] working non-peeler

[Paul Wagland]

<20040617091314.GA28476 at ebed.etf.cuni.cz>
<20040617141946.GA18688 at naranek.org>
Message-ID: <0C459B15-C0A5-11D8-A9B7-000A95CD704C at kungfoocoder.org>


On Jun 17, 2004, at 16:19, andrew lattis wrote:

> On 2004/06/17 11:13:14AM +0200, Thu, Petr Tomasek wrote:
>>> basically it take an existing ogg, decodes it to a wav, grabs the
>>> comments and re-encodes at a new quality setting and appends the
>>> comments.
>>
>> Just curious question: will there be some quality loss?
>>
>
> i don't know enough about the encoding to say 100% but i wouldn't think
> so, in theory the wav should contain all the information that was in
> the
> original ogg, so your only loss will be whatever would have been cut
> out
> in going to the lower quality, i haven't been able to hear any defects
> atleast.

This is not strictly speaking true... this can be trivially tested by
just doing a ogg-encode/decode cycle on a given file for about 10
times... You will notice the difference, particularly at lower
bitrates. The problem is that what you get back is not what you put in
"minus some", it is a reconstruction that sounds similar. The
distinction is important, this would not be true if we were able to use
real real numbers (float and double don't cut it ;-) since the
reconstruction could pick up exactly the same information, but since we
are decoding into a discrete space, then our nice smooth curves aren't
anymore... and that is what makes it hard to re-encode them.

Cheers,
Paul
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
Url : http://westfish.xiph.org/pipermail/vorbis/attachments/20040617/eb4149e3/PGP.pgp