[vorbis-dev] From LAME mailing list

[Monty]

On Thu, Apr 26, 2001 at 12:04:36PM +0200, Aleksandar Dovnikovic wrote:
> Comments?
> 
> ----------------------------------------
> "Mark Taylor" <mt at sulaco.org> wrote:
> 
> [...]
> > This is related to one minor objection I have to vector quantization
> > based codecs like Vorbis and the MPEG4 VQ codec: they do not compute
> > the quantization noise during the encoding process.  The choice of
> > codebooks (use a big codebook: low quantization noise, use a small
> > codebook: large quantization noise)

No, the vorbis codebooks only affect dynamic range.  The spacing of
codebook values in the current encoder (and for the forseeable future)
is a fixed and known quantity across all books.  Noise is not computed
because noise is a known quantity at all times, and is a function of
the floor value, not the codebook chosen.

> > If for some
> > reason there happens to be a lot of quantization noise in the 3khz
> > region, the codec is not aware of this and is unable to move bits
> > around to correct it.

That's not correct.  The concept of psychoacoustic noise in vorbis is
different from mp3, but it does know how much noise it considers
acceptable in a specific band, and how far off the actually acheived
s/n is.  It's not computed explicitly because it doesn't need to be.

(It is correct to note that in b4 this information isn't actually
used.  In rc1, it will be.  Vorbis doesn't move bits around because
the psychoacoustics in Vorbis are not iteratively refined.  Every band
gets exactly the number of bits it requests first time around as a
feature, not the lack thereof.)

Monty

--- >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.