[Vorbis] Bit rates, files size

[Graham Mitchell]

> Actually, the file sizes ARE different... I have tested it, using
> a single WAV source (Mozart´s 25th symphony in D Minor, if that is of any 
> interest...). MP3 files are smaller than wma, and ogg smaller than 
> both the others. I used dBpower AMP, and it indicated the supposed 
> bit rate for the ogg files. All conversions using CBR, even for ogg; 
> the VBR files were smaller still, but I guess that was expected.

Ogg doesn't do CBR.  The best it'll do is a clamped VBR which tries to average
the bitrate over a certain period of time.  And with VBR, if it needs less
bits to encode a section that your target "CBR" bitrate, it doesn't use them.
 This is especially common in classical music, where large chunks are fairly
easy to encode because only a couple of instruments are playing.

"CBR" mp3 is especially stupid in this regard, as it'll use (for example) 128
kbits to encode a particular second of audio even if 90 kbits is more than
enough to reproduce the sound.  Vorbis (and probably WMA) will both use the
fewer bits if that's all they need to reach a certain quality.

Did you read the rest of my "Introduction to Compressed Audio with Ogg
Vorbis"?  All the questions you're asking are answered in it.

http://grahammitchell.com/writings/vorbis_intro.html

> Lets think of something like creating different music files, and
> comparing them back to the original DIGITALLY; can you calculate in
> any way which file is closer to the original?

Yes, this isn't very difficult to do.  Your suggested scheme would work fine,
as far as I can tell.

> BTW, is this a good indication of sound quality?

No, it isn't.  It isn't a good indication of sound quality at all.  That's the
whole IDEA behind psychoacoustic modeling for lossy compression.  The whole
idea is that even though the ones and zeros AREN'T the same, they SOUND the
same to humans.  And if you come up with a scheme to model which things SOUND
the same, then you've basically created your own perceptual audio codec (as
Haxe mentioned).  Which is why Vorbis sounds better at a given filesize; the
bits are MORE DIFFERENT with Vorbis audio than mp3 audio (and so can be
compressed smaller), but they sound JUST AS GOOD.

The ONLY good test for the quality of a perceptual audio codec is your ears. 
Or rather, many trained ears, with good equipment, in a double-blind listening
test.  The most recent of these, Vorbis came out on top (roughly tied with
MPC, which is non-Free).  Lame-encoded mp3s (the best mp3 encoded out there)
came in 4th out of 6, and WMA standard was 5th.

http://www.rjamorim.com/test/multiformat128/results.html

-- 
Graham Mitchell - computer science teacher, Leander High School