Hi,
I found the latest Vorbis codec by aoTuV and Archer, to be *perceptualy* transparent at q2 which is around 96Kbps.
The software codec is available here:
http://homepage3.nifty.com/blacksword/libv...3_ArcherB10.zip

Then I went furder and downloaded some of the hardest content for Vorbis, from the Roberto's public listening tests page.
For several of his test the samples "Kraftwerk" and "Waiting" has been one of the worst for Vorbis.

But I just downloaded the Flac files and recompressed in WMA Q50 VBR version 9.1 and OGG vorbis at q2 using the previously mentioned coded.
To my surprise this sounds very good to me, it was hard to hear any difference with the oroginal Flac files. With the WMAs it was a different story, since I could hear some sound quality degradation.

So I wonder why people here in this forum are saying that Vorbis is transparent at q6 or more !?!?
To me is excellent at q2 !!!

If you want to hear for yourself, here are the original test samples.

http://www.rjamorim.com/test/samples/kraftwerk.flac
and
http://www.rjamorim.com/test/samples/Waiting.flac

Any comments appreciated.
Thanks,
Lucem

It just means that there are people with better hearing and audio equipment than you.

Could you give us a little more info on your listening environment? Transparent at 96kbps is a bold claim..... not a bad one.... just a bold one.

Happy New Year btw.

You might just not be trained at detecting artifacts (knowing what to listen for).
But I agree that the latest AoTuV beta gives impressive results at such bitrates, I'd never call them transparent though.

Ok I just made some test in some hard to encode music samples.
You guys can download the files packed in a zip file here (3MB)
http://www.kilobox.com/stuff/audiopack.zip
The samples are compressed in OGG q2, MP3 Lame Engine 3.96 using V7 and MS WMA Q50 VBR version 9.1.
The goal was to make files of equal size aound 96Kbps.
To me the undisputed king is OGG Vorbis, followed by WMA, at those bit rates Lame had to use 32Khz.

My listening setup is a Custom made PC with Asus Motherboard, a Turtle Beach Santa Cruz sound card and a "closed ear" high end Sony headphones.
It would be interesting to hear some music samples where artifacts show up using Vorbis at q2.
I stilll have to find one.....

i did some ABX tests with ogg vorbis myself, to see at what q level it becomes transparent to my ears and i always could ABX q3, but not q4.
so i encoded a few albums at q4 and found one song where i could ABX it, so now i'm encoding at q5 and i'm still looking out for artifacts, but so far i've found none.
i don't think my hearing is that good, so it surpises me a little bit that q2 is transparent for your ears.

-andy-

To original poster:

You aren't alone. I find Vorbis transparent at q3 most of the time.

Not all music is easily noticable at having artifacts at that bitrate. Another thing alot of these folks have trained themselves to know what to look for, it is almost second nature.

It is like with MP3Pro, I can always pick it out. Not because any artifacts either, but because the SBR part of it is lower volume than the original(if not the SBR part, then very near it because it is the higher frequencies that have lower gain than the original). It could be just a flatness that you aren't even aware of or an ambiance that is common with the codec.

Anyway, if you try to hard you will find what you are looking for. You may not like what ya find in the end though, so be cautious of wanting to know to much.

I find Vorbis transparent at -q 5 and very listenable at -q 3 and 4. Not everyone thinks -q 6 is the minimum but I suppose some people can hear stereo seperation problems better than others along with other artifacts better than others. I wouldn't worry about it. If -q 2 is good enough for you then that's just how it is

The point is that ogg is not a transparent codec at any bit-rate, if it was there would be no problem samples, no one would be able to ABX differences from an original source file & an ogg encoded file.

Not being able to ABX known problem samples or hear differences at low bitrates simply reflects on peoples equipment &/or hearing, it doesn’t make a codec any better.

That doesn't make sense. Transparency is subjective. If someone doesn't hear a difference, then the codec's transparent for that person, with that sample, at that setting.

It would certainly be a transparent codec at that specific bitrate.

edit: Q3 oggs are transparent to me almost all the time. 128kbps MP3 is not transparent a lot of the time. That makes Vorbis a better codec for me than MP3.

Ok. What I mean by transpanrent is PERCEPTUALLY transparent.
Since OGG is a lossy compressor, like many others.

Bottom line:
Can some one here dare to back his/her claims with some samples using the mentioned OGG version in q2 ?????


I'd like to HEAR IT, please.

Thanks,
Lucem

Of course you mean perceptual transparency.

About problem samples with Q2, there's no dare to it. If you really want to know what the fuss is about, search for some problem samples and try listening to Q-1 oggs and the originals.

You might be able to notice something similar in Q2 files too, but you might not want to do this if you don't want to get accustomed to artifacts.

But then, there are people like me who have had experience with low-bitrate artifacts but are comfortable with Q3..

Try good old castanets.wav.
OggEnc v1.1 (ArcherB10 based on AoTuV Beta03) at quality 2

I have recently decided to move away from mpc because of portable support. So the choice was either mp3 or Vorbis. I've decided to go with Vorbis at Q6. Gives comparable bitrates to mpc xtreme and mp3 aps.
But soon I had to find out that Vorbis at Q6 is far from transparent. Most noticeably the noise, high frequencies problems that I experienced some time ago when I tested Vorbis are still there. I have uploaded a test sample, if anybody is interested here.
Btw, I can't abx mp3 aps on that sample.

So after all those years, going from mp3, trying different formats, the circle closes and I'm back to mp3 (aps to be exact).

Edit: The used Vorbis encoder was the official 1.1 version.

Ok,
I just compress some test samples from here:
http://www.soundexpert.info/theory.htm

And quite frankly the only sample that I can hear some problems is with the French Male Speech.
http://www.kilobox.com/stuff/oggq2/frmalspeech.ogg

with q-1 is very obvious.
http://www.kilobox.com/stuff/oggq-1/frmalspeech.ogg

...And Castanets at q2 was very very close to the original.
http://www.kilobox.com/stuff/oggq2/castanets.ogg

Now folks, we have to put this thing in perspective.
These are Lab samples and no real music.
I think with 'normal' music content compressed with ogg q2 is very good almost transparent to me.

Of course I would always keep a higher quality copy like MP3 256Kbps or Flac, as the original source, but for portable Audio players I think ogg q2 sounds fantastic, like 96% transparent.

The same thing happened to me... but til today I use all three presets, depending on what I think the encoded kind of music "deserves".

Well, some people (especially around this forum) are more sensitive to artifacts. If q2 sounds good to you then use it and enjoy the enormous amounts of music you can store. q2 is not bad in any way; it is very efficient imo.

Once you get accquainted with the sound of artifacts you start noticing them in regular music as well. I was surprised that very often what I thought were easy to encode samples turned out to be quite difficult for this or that codec. On the other hand there is lots of electronica like Aphex Twin or Autechre which is just full of clicks and pops which normally would be called "lab samples" but that would disqualify a whole genre of music. If you examine other hard to encode samples most of them are just normal music. Only a few are of the weird electronica type.

Q2 can sound quite nice, especially for the size, but I've developed a sour taste for Vorbis's claustrophobic and occasionally odd stereo imaging. Even good ol' quality 4 with its best bang-for-the-buck reputation sometimes suffers from unnecessary stereo collapse.

If I had it my way, I'd combine quality 4 with lossless stereo coupling (or maybe the point stereo settings from quality 5). It's a shame the Vorbis encoder doesn't have the myriad of switches available like LAME and Musepack.

I'm sure that can be arranged

Why do I get the feeling I'll get another encoder from QK to play with?

Unfortunately, I don't have much free time to do it now so it will be a little while before I work on it. But from what I've seen (from adding the ITP switch), it doesn't look very hard to do. It's just working out which psymodel values to tweak.

I envy you...I find the castanets sample to be ghastly at this bitrate. Pre-echo, coarseness, and HF noise/hiss "pumping" is readily apparent. However, I've beat this particular sample to death with every codec out there and know exactly what to listen for in an ABX test. So, be careful about how low you set your quality level, because the first time you notice something "wrong" with an encoding, you will hear it (and possibly more issues) better every time you listen. That's why I would personally advise against using anything less than -q4 for encoding, save for your portable device. I don't know about you, but re-ripping CD's is about as much fun as a colonoscopy. (Not that I'd know, of course )

(Edit: typo)

There are likely a variety of variables to tweak. For instance, LAME has the "nsmsfix" variable which is a single number that controls the mid-side stereo switching threshold...and "shortthreshold" which is like Vorbis's impulse_trigger_profile but is a decimal value rather than a preset integer. It would be nice to play around with the masking levels too. Just to see what this puppy can deliver (not that Vorbis isn't tuned but rather to adjust the model to our individual ears and preferences...squeeze a little more bang out of the buck).

Cool, thanks for the info. I'll look into that when I have time.

The impulse_noisetune switch plays directly with the noise masking levels for short blocks (adds an offset). Finer grained control will be a bit difficult since there are lots of values to change per quality level.