[vorbis] APPLAUD.WAV problems

[Ed Sweetman]

On Sun, 2002-02-17 at 18:27, Segher Boessenkool wrote:
> > That's why I said it's probably done by effect and not actual straight
> > forward coding it in.
> 
> Because there's a minimum and a maximum "complexity" / "entropy" in your
> test sample.
> 
> > the average should depend on the complexity of the clip,
> 
> It does.
> 
> > the average should depend on the complexity of the clip, the -q setting
> > should not be setting an abr.
> 
> It doesn't.
> 
> > I thought we got rid of that abr crap for real vbr.
> 
> It has never been like that.

Yes it did. When you would set the bitrate you're trying to get and the
encoder would try to match it.  It wouldn't be exact but it would be
near it.  It was probably around beta4 or so I forget, but if it wasn't
abr then developers did a damn good job at making it look like abr.  I
even believe at one time there was an abr and vbr setting for oggenc, by
not supplying an abr it would default to some vbr value.Again, this
could have been even earlier in the beta series. I've been using ogg
since it came out so I dont remember the exact version of it. 

<p>> > Now assuming there is some low inaudible noise in a silent
> > clip, the encoder should be stripping out inaudible information anyway.
> 
> How do you know it is inaudible?  You don't know the setting of my volume
> knob.

Well, we've got silence with noise in the human hearing range that's
just so small you cant hear at normal volumes and then you have silence
with noise outside the human range.  Of course the encoder should be
tearing the later out. 

The former can still be solved by some advancement of the psychoacoustic
engine. It has to be because you can set the quality setting to 1 and it
sounds exactly the same as when you set it to 5 or 10 or whatever. Now
your volume argument if it applies to this must also apply to normal
sound.  So if you can hear differences in the silence at different
bitrates then you can hear the differences of tones and music at all the
different bitrates at a high enough volume,  but how much are we willing
to magnify the audio to see those differences before we decide that the
audio is too insignicant to play a role ?  

Rather, more importantly if the psychoacoustic engine can remove audio
that it deems insignificant from a normal audio file, then it should be
able to tell that the audio in a clip of noisy silence can be encoded
with a rather insignificant amount of information.  

Instead you seem to be hinting that the encoder doesn't care about the
amplitude of the wave when despite your volume argument I think it
should care about the amplitude of the wave in relation to some constant
of human hearing response. Sure, be a little generous about what the
human ear is capable of discerning from silence, but you cant say just
because the surrounding waves are the sign amplitude then you have to
treat them all like it's the same kind of audio file as a cd clip.  If
that's reason not to take information away from it then you would be
able to take even less information from normal clips because it's
changes would be audible at less high volumes than the silence.  

I'm really not sure if i'm making this clear for you. The psychoacoustic
engine should be able to say, hey, if the audio being encoded is within
certain constraints after being decoded then we should use that
bitrate.  In this case an extremely low bitrate for my "silence" would
be possible.   Basically what you're saying is that the encoder works on
the audio as if it were normalized and I'm saying psychoacoustics should
be able to determine a sufficiently low pattern of audio to be silence
for all lossy formats and only keep that sufficiently low pattern of
audio if it's being encoded at a lossless quality.

Why should the ogg encoder have to take into acount someone not
normalizing their audio ?   

<p>> > Again we come to a problem with the psychoacoustic engine again I
> > guess.  If things are really that bad then that could explain why
> > nothing is getting under certain bitrates at certain quality levels.
> 
> *Real* silence will always be a *really* low bitrate.
> 
> > can you explain why there are no low bitrates for audio sections that
> > require very little information to be encoded to have the quality that
> > is designated.
> 
> Because a) your assumptions are wrong or b) this makes the _average_
> bitrate lower.
> 
> 
> 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-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.

<p><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.