[vorbis] APPLAUD.WAV problems

[Monty]

>  > 2. From approx. -q 1 on the CHIRPS seem to fade out, but there is a
> *VERY*
>  > noticeable BOOSTING OF THE HIGH FREQUENCIES (or high frequency noise) >
> like if you boosted TREBLE with an equalizer. High frequency boosting has
>  > already been discussed on this forum
> Yes, high frequency noise. It is definitely there, up to "-q 4.99", even if
> you have to strain a bit to hear it at that level.

It's not a boost, it's a tendancy to concentrate low amplitude noise
in the center.  vorbis/doc/stereo.html discusses this.

This sample only has applause and noise to the sides and practically
none in the center.  Point stereo moves noise to the center where it
becomes easy to localize.  There's no great mystery here :-)

> Yes, it disappears completely at "-q 5": that's where the lossless stereo
> encoding kicks in. It looks like the lossy stereo encoding cannot encode the
> big differences between the two channels in this sample.

it's not intended to.  Even at q4/128, you have to make compromises.
Given the choice between strictly purist fidelity and a sample that
sounds better as a standalone sample, we opt for the latter whenever
there's a compromize to be made.  That is, 'sounds different when
compared to the original but still sounds good' is always better, in
my book, than 'sounds more like the original, but the artifacts are
obvious standing by itself.'

>  > LAME does a **MUCH** better job, even at 128 kbps (!!). This really
>  > confuses me, because I use OGG and consider it to be a better performer
>  > otherwise.

There's no one definition of 'better'.  LAME is sacrificing other
elements to maintain the stronger stereo image (namely, obvious
flanging).  I prefer oggenc's handling of the sample... but it's
probably not surprising that our Ogg encoder reflects my tuning
preferences.

You will always be able to find a sample at a compromize quality
setting like -q4 (bitrate like 128) that sounds relatively worse than
others.  It isn't necessarily the right decision to hurt quality in
100 samples to improve one.

> Gian Carlo, would you mind explaining how did you tune your encoders - at
> least to the guys that hold on this project? 

You assume erroneously that Gian Carlo is not a member of the project
or that we don't talk to him.  Gian Carlo's tuning results are also
folded into every new release.

> I believe that your suggestions
> would help OGG making one step ahead towards being the best lossy encoder
> available.

rc4 will allow user selection of stereo mode, which eliminates the
need to argue.  I will say though that the majority of users prefer
lossy stereo coupling (with the compromise that implies) over lossless
coupling and having fewer bits available to stabilize the high end. 

We're talking about perceptual compression and everyone perceives
differently.  The only changes we can really consider are ones that
are a net benefit when considering all users and all samples.  We
can't tune applaud.wav to the exclusion of all else.  I know that's
not what you're suggesting, it's just something to keep in mind.

Monty

<p>--- >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.