At q 4 and below, aoTuV is definitely the better encoder (since GT3b2 at q < 5 is similar to 1.0.1 which has been shown in the listening tests here to be the worst). Above q 4, things are a bit confusing. GT3b2 addresses pre-echo problems but has no effect on HF boost/hiss at q 5. I'm not too familiar with aoTuV (perhaps Aoyumi can answer this) but I think the HF problems are addressed at q > 4 and pre-echo handling has been tweaked with respect to 1.0.1

A merger appears the likely solution though after I merged my pre-echo tunings in QKTune beta 3.2 with aoTuV, results weren't as I had expected.

If no-one is working on a aoTuV+GT3b2 merger at the moment, I'm willing to do that.


EDIT: After having a peek at the source, looks like aoTuV does include changes that affect pre-echo (possibly at q 6, 7, 8, 9). So a merge with GT3b2 will cause conflict. I suggest someone test aoTuV at these higher q's to see how it performs

The changed part of aoTuV influences in all the bit rates.
And it will make it hard to be audible in quantization noise (include pre-echo).

Although the pre-echo tuning by the high bit rate may be effective, balance may be broken down conversely.
However, it may have worth of a challenge.

By the way, the present aoTuV is the experiment version and is still due to change.

[Postscript]
I'm sorry, it had mistaken. The measure against pre-echo was slightly made after q6 of aoTuV. I forgot this.
However, original post is a fact again.

There are also Debian packages available at RareWares.

Of course, people will only be able to reach it once the host goes back online

Why don't you, Aoyumi, Nyaochi and whoever else is interested start a Sourceforge project where you can work together? Then, it won't be needed to later come up with hackish merges of each branch's best tunings.

It seems QuantumKnot is (was?) interested:

http://www.hydrogenaudio.org/forums/index....howtopic=19131&

Hi !

Could someone provide ogg librairies (ogg.dll, vorbis.dll, vorbisenc.dll, vorbisfile.dll) from aoTuV, Modest Tuning beta 3, QKTb3.2, and GT3b2-HF reduction encoders ??

10x

QKTune beta 3.2: http://www.rarewares.org/quantumknot/vorbisqk32-dlls.zip
aoTuV: http://www.rarewares.org/quantumknot/aoTuV-dlls.zip

I'll compile the other two later when I have time

Ok...Thank you very much QK

The fact of changing dll improve the quality we listen musics or theses dll impoves just the quality at the encoding ?
For example, if we want to listen a file with winamp and the music is encoded by aoTuv or QKTune, the sound will be better if we update dll ?
Or even if I keept always the same dll, musics encoded by ogg vorbis (futurs versions) will be always compatible ?
An other example , in games likes UT2003 if I replace currents dll by theses news dll, the sound will be better ?

These dlls don't improve the quality of decoding. It is only encoding.

aoTuV beta2 was released.
Some problems should have improved.
This beta2 believes that I am better than an experiment version.
aoTuV page

Good news. Thanks alot for your work!

Wonderful ! Have you already make test to compare with the previous version of aoTuv and the other encoders ?

The bitrate of the files encoded by this version seems perfect !

This encoder will may chose for the multiformat test ?

It would be good for some people to do some listening tests first. The idea of including a 3rd party Vorbis tuning into the multiformat listening test is still up for debate.

I thought it was already decided that the previous aoTuV encoder would be used in the listening test? The decision was based on the Vorbis listening test result, wasn't it?

Roberto should have the appropriate encoder now:
http://www.hydrogenaudio.org/forums/index....ndpost&p=202430

Frankly I don't understand why this is still being debated. As you already hinted, if we want to use another encoder we should first conduct a new listening test. I don't think there's time for that now.

Why don't we stop discussing this and just stick to the encoder that won the Vorbis listening test?

because we want vorbis to behave as good as possible

someone could also critize that we didnt test the first alpha version of aotuv, people often simply assume that new versions are better (of course it can be questioned if this is correct always)

still i dont really have time to run such a test atm
did rjamorim already announce when the test will start? or does he already prepare the samples?

Next wednesday - hopefully

That's exactly why we shouldn't rush to get the newest encoder included. It first needs to be tested. There is always a risk that flaws got introduced.


That has nothing to do with my argument. Because of the importance of the encoder choice, the decision was made to perform a Vorbis listening test. The test is finished, the winning codec was chosen, everybody agreed. End of story.

If we discard this choice for a newly released encoder, then why perform the listening test in the first place? And since Roberto wants to start the test next week we don't have time to organize a new test.

Therefor, my recommendation is not to take any chances and stick with the winner of the listening test.

The all-inclusive comparison test with other encoders is omitted.
However, it tunes up so that it may become an original sound more with my ear closely.
An important thing is how you feel by testing this.


Since a new version regards me as better, I want to recommend this.
However, the opinion of other people is important.

Aoyumi's personal results correlate almost perfectly with the average results for the group (harashin, bond, guru, aoyumi, and nyaochi, with extra results on Waiting.wav for LoFiYo and Mac). Here are the averages for the group:

101CVS 3.02
AoTuV 3.87
Mtb3 3.26
QKTune 3.74
GT3_HFR 3.75

And here are Aoyumi's personal averages:

101CVS 3.3
AoTuV 4.2
Mtb3 3.5
QKTune 4.0
GT3_HFR 4.0

Correlation is 0.997, which is the highest of any individual who participated in the listening test. The other correlations were:

bond: 0.98
guru: 0.36
nyaochi: 0.94

So, if Aoyumi personally tested all 12 samples with the beta 2 version against the older version to form his opinion, then it is very likely that the group would agree with his opinion.

ff123

Is there something wrong with my results?


I agree with that.

But it's not clear that Aoyumi tested all 12 samples to form his opinion.

ff123

Edit: Oops, Harashin had scrolled off my Excel screen. His correlation was 0.91 with the group (his numbers had been included in the group averages in my first post, so I didn't actually leave him out).

Edit: pretty much anybody who tested all 12 samples except guru, is likely to produce results which the group can agree with. Such a test can be shorter, since only beta2 needs to be compared against the older version, but it should include all 12 samples.

So, what's the deal? Which codec should I use? It's not very clear from this theme. The freedom of choice is not, what I need. I wanna know, which codec showed best results. Is there such page ready? And is GTune 3 beta 2 + QKTune beta 3.2 EXPERIMENTAL going to be tested?

updated RareWares Debian Packages available now, too
(html updates to come later, as always unfortunately...)


later

Based on the Vorbis listening test, aoTuV won at q 4, so at this nominal bitrate, it is expected to give you the best quality that the Vorbis codec has to offer.

The quality of the new aoTuV (beta 2) is not so definite. Or perhaps I should say, not as definite as the previous aoTuV that was tested by at least 4 people.

With regards to the situation, I personally would love to see the new aoTuV be used in the multiformat test, if it can be reliably verified that it performs better than the previous aoTuV we tested. There is a risk that there is some regression and the last thing we want is for this to appear in the multiformat test.

However, on the other side of the coin, I did accept the point made by Roberto earlier that we shouldn't go about tuning Vorbis continually in order to improve its chances in the listening test up to the last minute. That would be unfair. Therefore I took the initiative and froze my version of QKTune beta 3.2 on that day and Nyaochi also did so for MTb2 the next day. So I already departed from that rule slightly when I agreed to include the experimental version of aoTuV just before the test and I'm not sure if its proper that I bend the rule again after the Vorbis listening test had completed and we'd all agreed to that version of aoTuV.

Not to place the ball in his court deliberately, but it is ultimately for Roberto to decide whether he will accept this new version of aoTuV (if after preliminary testing, it gets the nod from listeners), to be included in his test or continue to use the one we agreed on in this thread.

If the older version of AoTuV is to be used, it should be made available on Aoyumi's page. Right now, only the beta2 release is available.

ff123

I still have my own compile:

http://www.rarewares.org/quantumknot/oggenc-aotuv.exe

Hopefully it is bug free

If anyone has some spare time, by all means, do a test comparing this new aoTuV with the old one. Irrelevant to whether it does or doesn't make it to the multiformat test, we would still like to know whether it is the best. I'm already contemplating using this to rip my audio collection

I won't accept anything without some valid backing up. Otherwise, I'll be making a criticism target out of myself. I only accepted featuring auTuV because there was a listening test behind it I could point critics to.

It seems that aoTuVb2 is superior to its previous version on the same samples.
aotuv_results.zip

It's probably a pretty safe bet that the other 4 people who listened to all 12 samples will also find beta2 to be better on this sample set (except maybe guru whose correlation with the group wasn't very high).

What is the bitrate penalty for this improvement?

ff123

Surprisingly, seems the b2 tends to produce smaller files.(at least with the 12 samples) Therefore, IMO including the b2 shouldn't be unfair to the 128kbps features tests.

Excellent. Thank you, harashin and ff123 (and of course Aoyumi).

Sorry if I looked overly paranoid, but I just wanted to remind everybody not to submit the latest and greatest version of a codec without testing. But then again, Roberto would never accept this anyway.

With all this progress on Ogg Vorbis, it's hard to believe that, only a few months ago, we had the infamous "Is Vorbis dead"-thread.

Well, if someone else can come with results that corroborate b2 is better than b1, I will include it in my test.

Anyone up to the challenge? I would need these results by monday the latest.

Edit: Thank-you for testing, harashin.

My listening result: aoTuV beta2 seems to be better than the previous version. Particularly, preecho problem in Blackwater and FloorEssence samples, background noise problem in MidnightVoyage sample and crash cymbal sound in thear1 sample are improved. IMHO, it's OK to use atb2 for 128kbps instead of at20040402 if someone else does not notice severe regression.

Here's the result.

is aoTuV beta2 available as oggenc yet?

Check his site out.

thanx harashin

OK. Considering harashin's and nyaochi's results, I would reckon aoTuV b2 should be used instead of b1.

Everyone agree with me? Can we get things rolling?

I think so, yes. Just gotta make sure that version of oggenc doesn't have the localisation difference with the decimal. If so, I can provide a binary.

I think I will go with 4 anyway. So, no worries.

But, of course, a version with . or , hard coded would be welcome

Oh yeah, q 4. Silly me

Well, here is my hard coded version, for anyone interested. It uses a . (dot)

http://www.rarewares.org/quantumknot/oggenc-aotuvb2.exe

Thank you for carrying out harashin, nyaochi, and a comparison test.

Locale fix oggenc (.version) of beta2 can be used from a test page.
This outputs the same data as a reference binary.
aoTuV test page

Moreover, on the page of aoTuV, oggenc and dll of the usual version (for win32) by offer of nyaochi are also prepared.

Now that we have some sort of scientific proof that aoTuV is the best Vorbis tuning out of the bunch, do you think it is time to make it the recommended encoder or do we need more rigorous testing before we retire GT3b2?

I think so. We, however, can recommend aoTuVb2 at -q4 and below at the moment.

So let's test aoTuV b2 at 180 kbps nominal bitrate (Q6), so we could compare it with Q5 of GTb2!

Comparing two codecs at such high bitrates is pretty difficult.

But in this specific case, it's maybe easier. I explain.
There are two known problems with vorbis at this bitrate. By "known", I mean generic problems, occuring with a lot of samples :
- noise issue (very small I admit after -q6): present with GT3b2, maybe removed with aoTuV
- pre-echo issues (very high with CVS code): excellent correction with GT3b (but maybe excessive bitrate inflation), aoTuV need to be tested.

By focusing the tests on these two problems (using percussive samples for pre-echo, and quiet samples for the noise problem), I suppose that we could build a serious opinion about respective quality of these two vorbis encoders.

I was far from internet these two weeks, and my latest aoTuV encoder was the prebeta2 release. I've played a bit with this one, mainly at 96-128 kbps. I've also encoded some samples at -q6, and noticed sometimes additionnal noise, not very hard to ABX. IIRC [I don't have access now on my main computer], on one sample the noise was audible up to -q7. Obviously, the prebeta2 didn't solve the (or maybe "a") noise issue at high bitrate settings.

Testing is possible, and it's worth to try.