[vorbis] RC3: I'm impressed

[Nicola Larosa]

SUMMARY

I'm impressed.

RC3 withstood anything I could throw at it. "-q 3" is really Vorbis' sweet 
spot, almost perfect, and "-q 0" is eminently usable, with only marginal 
defects for normal usage.

Already sent money, will do that again.

* * *

I spent at least 40 hours in the last week testing RC3. I selected a few 
fragments from pop, classic and jazz CDs in my collection, covering piano, 
harpsichord, voice, percussions, cymbals and complex arrangements. Also, 
downloaded a few samples from the Net.

I listened to them with a decent Sony headphone, and looked at them through 
the spectrograms of CoolEdit2000 (have yet to find a good audio editor on 
Linux). I was able to see many more differences than to hear them, but it 
was sometimes useful to loop a short fragment and hear it over and over, to 
"learn" it and focus on its qualities.

I worked hard, but found little to complain about. RC3 is *so* transparent. 
Did not even think about ABXing, since I mostly could not find defects even 
when knowing I was listening to the encoded signal.

Most evident differencies between original and encoded are to be found 
through synthetic test signals. A good one is EncoderHell from 
AirWindows.com . It was the only case where I could see (not hear) some 
pre-echos, and some differencies between "-q 3" and the upper quality levels.

Another difficult signal is the sine wave (1 KHz), that is a little wobbly 
at "-q 0", but inaudibly so at "-q 3".

The only difference I could find between the original and the "-q 3" encoded 
was a static section of a complex arrangement (cannot recall which sample 
:^( ), that sounded a little different, maybe more "metallic", but only an 
immediate comparison could evidence this. The difference lessened going to 
"-q 5".

The omnipresent cut at 16 KHz in "-q 0" did not seem to bother more than 
some. At that quality, "applaud.wav" and "castanets.wav" rendered 
transparently. There are some artifacts in "fatboy.wav" when the voice goes 
low, but they disappered completely at "-q 3". In "velvet.wav" the hi-hat 
was a little flanged, but got OK at "-q 1".

And that's it. The piano, the sax, the harpsichord, the voices, I could not 
detect the difference from the original, even at "-q 0". Another thing, I 
could not evidence the difference between point stereo at "-q 4.99" and full 
stereo at "-q 5". Is there any sample around to hear that?

My temporary conclusion is that Ogg Vorbis RC3 gets archival quality at "-q 
4.99" (around 120 kb/s), just to err on the side of safeness, even if "-q 3" 
(around 105 kb/s) would probably be enough in most situations, and "-q 0" 
(around 60 kb/s!) is eminently usable when memory is tight, such as in 
portable players (when we'll see them :^) ).

And I can't believe I'm really saying this.

If Monty can work this kind of magic for lower bitrates (and sampling 
frequencies) too, I cannot see any competition from any proprietary 
streaming codec, Real or Imaginary. ;^)

I did not test any previous version. From Monty's comments to the WayItIs 
sample in the listening test at ff123.net, and from other comparisons in the 
last months, I gather that there's been a lot of optimization done to this 
release.

I understand that there's a lot of people working on Ogg and Vorbis, and I 
really appreciate the work of you all. It looks like the core codec work is 
done by Monty, and the quality of this result by just one person says 
something about that person. Also, I really like how he answers questions in 
interviews, and on Slashdot. :^)

I know that MPEGPlus too is done by one person, and seems to work well, but 
I did not, and will not, test it, since its author is not willing to open 
source it, and this renders it useless to me.

So, hats off to Monty, and to the rest of you! I already contributed via 
PayPal a few weeks ago and will do so again soon. I think it's important to 
concretely show our support to this group, and facilitate their work in any 
way we can.

I'm left with just one doubt: which is better usage of my time, doing 
further testing with Vorbis or translating that Perl ABX script to Python? ;^)))

<p>
-- 
"Mozilla will be around long after nobody can remember
just quite what Internet Explorer actually used to be."
   AirLace on Slashdot

Nicola Larosa - nico at tekNico.net

<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.