[vorbis] APPLAUD.WAV problems

Ed Sweetman ed.sweetman at wmich.edu
Sun Feb 17 13:39:13 PST 2002



It seems to me that something that encodes a difficult audio sample at
lossless quality and easier things at lossy quality is just what VBR
should do at higher quality levels anyway. So I dont see the whole
problem here. Perhaps you're looking for a more aggressive high quality
VBR level with a size bias instead of a lossless bias.  

So the vbr routine would encode as low as a bitrate as the designers
decide is suitable to be still undetectable but gives the VBR routine
the option to encode a sample at lossless bitrates for things it thinks
just cant fit into that undetectable range easily.  

It was my understanding that vbr already does this as it has no real
upper or lower bounds, the quality level just makes things more and more
aggressive as your quality level goes lower and lower and less
aggressive as it gets higher.  

The reason applaud.wav is a problem (if it is at all anyway) is that
there is an upper and/or lower bound bitrate setting for quality modes
for vbr encoding and that would indeed be a problem, otherwise i'd have
to say it's an issue with the vbr routine itself not being able to
correctly assign the required bitrate. 

my initial tests show that higher quality settings put a lower bound,
meaning no bitrate will be assigned below a set bitrate at a certain
quality setting.  At lower quality settings (below 4-5) you'd probabyl
see the opposite occuring.  Of course this is going to be a problem if
someone has a file with only small portions needing extremely high
bitrates to encode the file and they want the smallest size possible
while still being listenable.  

I dont know enough about acoustic algorithms and such to be able to
suggest any changes to the VBR routine, but if it's possible to instead
of fixing the characteristic of setting lower and upper bounds based on
-q settings, have it set the aggressiveness of achieving lower bitrates
with -1 being the most extremely aggressive and 10 being lossless and
having another setting to override it's default policy of keeping the
audio without any audible artifacts.  such as  -q 1 -b size,  will cause
oggenc to ignore it's default policy of keeping artifacts out by setting
an upper limit denoted by the quality setting set,   1 would be a low
upper bound like at around 64-92 etc. . this can be debated over
later.   

What I'm saying is the vbr's routine policy should be no artifacts at
all costs, meaning setting your file to encode at -q 1 and not setting
the bias flag to size will encode your file with the most aggressive
routines, cutting corners more than any other quality setting but still
produce a file that is lacking annoying artifacts.  meaning if you use
that setting on a complicated file, you'll get a pretty highbitrate
despite setting it to 1. If a file cant encode below say 192kbit without
having annoying artifacts, then it wont encode under 192kbit no matter
how low of a quality setting you go at.  unless you specify the size
bias flag.  

This is not to say that every file will be around quality 5, I'm not
saying to make the vbr routine produce only cd quality clips, just it
only produces clips that are lacking any kind of artifacts that are
audible. Files will get granier and more radio like as you get more
aggressive but you will be guaranteed to get a clip that is as small as
it can be without having blips and flangs and such. I think that's how
vorbis's default vbr policy should be, then people who would like to
sacrifice that guarantee can put a bias on size and the routine can
forget about being held back those constraints.  

<p><p>just my 2c.  

<p><p><p><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.



More information about the Vorbis mailing list