[vorbis-dev] Thought for the new year

[Segher Boessenkool]

Gregory Maxwell wrote:
> 
> On Wed, Dec 27, 2000 at 04:41:55AM +0100, Segher Boessenkool wrote:
> > 1) MDCT is good for image coding
> > 2) image coding and audio coding are two very different things
> > 3) combine 1 and 2
> 
> It follows that MDCT is good for many differnt tasks. No? :)

"good" is subjective ;-) Hey, I said for the new year, you're too
early!

> > 4) if a psycho model is good, after leaving out what it tells you
> >    you can without hurting quality, applying the same model should
> >    yield the same results as you got before
> > 5) from 4: decode -> encode -> decode should result in (almost) the
> >    same stream.
> 
> Only if the system is stable. With most MP3 encoders, the stupid subbanded bit

Even if unstable: "almost" the sam results.

> allocation routiene is greatly affected by the quantization noise from prior
> encoding runs. Also, things like the mid-side selection may be triggered by
> quantization noise.

Don't quantize by rounding to nearest integer; don't do M/S <-> L/R selection
based on only high-frequency energy.

> A while back, I did some testing and found Vorbis to be fairly stable
> through multiple encode iterations, more so then the mp3 encoders I tested.

This is great news! I never did any such tests of Vorbis; but the psycha
isn't good in this respect, I'm afraid. It would be worse, of course, to
artificially fit it on it, as that's not the purpose of the psycha model.

> I suspect that current and future fine tuning of things like block switching
> might, however, make the encoder less stable in this regard.

That's the point I tried to make: if, for example, the encoder decisions about
block switching are based on something you can't hear, there's something
fundamentally wrong. I'm not saying it is now, mind you :-)

> Whats more intresting is that the framing used in vorbis will allow many
> types of editing without a complete decode/recode cycle.

I was trying to point out that a _really good_ psycha model would give
the same result after encoding->decoding, just like taking a picture of
a photograph should look the same as the original photograph.

Dagdag,

Segher

--- >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-dev-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.