[vorbis] Is Vorbis really THIS good?

[safemode]

> 11. oktober 2001, 23:08:13, you wrote:
>
> s> Would be nice to see oggenc or some other ogg encoder with the abiltiy
> to s> encode at whatever bitrate is necessary for the ogg to sound good.
> Instead
> s> of guess and checking that is necessary now (which slows down encoding
> ogg
> s> files for cds quite a lot unless you dont care about space/quality
> s> tradeoffs).  There is no one bitrate fits all.  Lame accomplishes this
> with
> s> mp3.
>
> >You mean something like real, quality based VBR, and not current
> >kindof-ABR?
>
> The current encoding modes are really VBR, but the way things work
> internally hides this from the user mostly.
>
> For example, when playing Oggs with Winamp it seems as there are
> only very small fluctuations in bitrate. If you try ogg123 you will
> see that they are actually much bigger.
>
> Another thing is that it looks as if you encode by bitrate, but
> that isn't actually true. Internally the current libs don't care
> about bitrate at all, and use preset encoding modes based on quality
> levels.
>
> Because Vorbis bit allocation is much more advanced than MP3, and
> the preset modes in RC2 are so stable, those quality levels usually
> turn out to average around a certain bitrate, which is why they appear
> ABR, but they really aren't.

What i'm seeing has nothing to do with advanced bit allocation.  In fact as 
it stands now, it's quite inferior to lame's VBR encoding.  If i want a high 
quality ogg file, libvorbisenc encodes everything near the bitrate i set 
(because there is no quality argument) including areas of the song that dont 
need to be just so the ABR comes out around the one i chose.  

> When you compare to MP3, keep in mind that, for example, to encode
> frequencies above 16kHz, MP3 encoding gets very varying bitrate
> requirements (has to do with a limitation in the format, you can
> read about it on the LAME mailinglist or r3mix.net forums), which
> causes the wide bitrate fluctuations you see. Vorbis has no such
> limitation, and can do with a lot smaller fluctuations.

my program, mp3stat, seems to show that ogg vbr is confined to the few 
bitrates around the bitrate you chose.  The reason why vorbisenc is still ABR 
is because if it was true vbr, I would see a file that has very little 
information encoded at a low bitrate, despite setting it for high quality (-b 
192).  For instance.  If i encode an mp3 at high quality.  I will still see 
extremely low bitrates in places in the file with mp3stat that required 
little info to be encoded.  With ogg i see ABR patterns.  Almost no bitrates 
are used more than 2 places away from the bitrate chosen at encode time.  
This is not true VBR, this is most definitely ABR.  
Flunctuations are not bad.  If the encoder can encode the audio at an 
extremely low bitrate and get away with it, it should.  It should not be 
confined to an ABR that you give it on encode time.  

> My mode (and possibly some of the modes in RC3, from what I've heard)
> is a bit more aggressive in certain circumstances, which cause
> wider fluctuations, and hence appears more VBR-like.

I hope rc3 has some different encoder flags that make more sense than just an 
ABR choice.  the encoder should decide what ABR is needed depending on if you 
want the file to be size optimized or quality optimized.  

> RC3 will also feature a bitrate management engine, which makes
> CBR encoding possible, and probably other interesting CBR/ABR/VBR
> tunings.

my little program, mp3stat is at http://safemode.homeip.net     I would 
suggest using it with gtk's default font set to something small but it'll 
work regardless ..you'll just have to scroll the text boxes.  2.9.5 sets it's 
own font.  It gives you a fast way to compare an ogg bitstream to a mp3 
bitstream or two oggs encoded at different settings   etc etc.    I find it 
semi-usefull anyway.

--- >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.