I just started a new public listening test.
Addition: a specific web site for the test has been created: http://rarewares.hydrogenaudio.org/test/

The purpose is to find what encoder performs best encoding AAC at 128kbps

The encoders and parameters tested are:
Psytel AACenc 2.15 -br 128
Ahead/Nero 5.5.10.35 128kbps CBR, high quality
Sorenson Squeeze 3.5 (FhG Pro) 128kbps
Apple QuickTime 6.3 (Dolby) 128kbps high quality
FAAC 1.17b -a 64 (64kbps/channel)

--Who should take the test?

Anyone interested in AAC quality, or people that have no interest but would like to help making this test better.
You don't need excellent hearing, but some good gear is welcome (headphones are must-have).

--How do I take the test?

Download ABC/HR + the readme file here or here. If you already have ABC/HR, get the readme here.
Further instructions, and links to the sample packages, are inside the readme.

For people using BitTorrent: I strongly suggest you download using it.
The torrents are available here:
http://static.hydrogenaudio.org/extra/AACTest/
It would help me save bandwidth and you will probably download faster too

The average size of the sample packages is 4Mb. It sums to 47Mb, all the 12 packages.

Edit: For those that already downloaded packages, read here

--I'm on Linux/MacOS/something other than Windows. Can I take the test?

You can. But you won't be able to use the ABC/HR utility, that is Win32 only. Since the reference file must be hidden (That's the HR in the name), tests performed without it wouldn't be valid. So, I can't plot your scores against others. But your comments on sample quality and artifacts would still be welcome.

--When will the test finish?

The test is scheduled to end on June 15. This date may be extended if proved necessary.

If you have any other question, please post in this thread.

Thank-you very much,

Roberto.

Just something I think should be clarified:

Don't post your test results before the test is finished

For several reasons, that practice would be harmful to the test.

Thanks for understanding.

Why are links nearly invisible? I setup Mozilla to display links as underlined blue text, but in HA it is only underlined and
gray like the rest of the text.

Here are the full links, if someone is having problem with the "here"s

ABC/HR: http://pessoal.onda.com.br/rjamorim/abc-hr.zip or
http://audio.ciara.us/rarewares/abc-hr.zip

Readme: http://pessoal.onda.com.br/rjamorim/Readme.txt

.torrent's of all the files:

http://static.hydrogenaudio.org/extra/AACTest/

A BitTorrent client is really a simple plugin application that enables your Windows, OS/X, or Unix/Linux machine to participate in swarms. These swarms of traffic facilitate file sharing.

The simplest .torrent client for Windows:
http://prdownloads.sourceforge.net/bittorr....1.exe?download
Just install it and you should be able to download from the torrent links using your browser.

More advanced client:
http://btplusplus.sourceforge.net/

Oh, and if you want to do this and help share the bandwidth burden, please leave the download client open after it finishes. This creates a torrent "seed" which is what helps to propogate the file throughout the swarm. The more seeds, the faster everyone downloads.

If you want to create a perminent "mirror" of the files without having 12 seperate download boxes open, you can either download: http://btplusplus.sourceforge.net/ and point the incoming dir to where your .torrents are, then attempt to start downloading them again (overwritting the old ones), and this will create seeds and share the files. The other option is to download the bittorrent sources ( http://bitconjurer.org/BitTorrent/download.html ), making sure to have python installed on your system, and launch "btlaunchmany.py <somedir>" or "btlaunchmanycurses.py <somedir>" on the dir with all the .torrents.

Any way to provide a random link generator (of the 13 .torrent's) so that all the samples get an equal number of tests?

Also, are these .torrent's the only way to download the files right now?

I downloaded BT++ v0.5.4, set the Incoming dir to where the .torrent's already are, did Add Torrent -> from URL (and pasted the ha.org url), but nothing seemed to happen when I clicked OK. Nothing listed under Transfers or History. Will this program work properly if I'm behind a firewall, or do I have to open some ports?

Unfortunately, I think it's too late now to come with such solution. Hopefully, people will pick up samples randomly, and whenever possible pick them all.

BTW, there are 12 torrents.



The links to download the .zip files form a server are inside the readme file.

You need to open some ports probably (from the faq: http://bitconjurer.org/BitTorrent/FAQ.html ):



For more info you might try:

http://www.filesoup.com/faq.html

or

http://smiler.no-ip.org/BT/BTFAQ.php

or

http://smiler.no-ip.org/BT/BTTutorial.php

Whoops, too many programming assignments... I start all my counts at zero now.


It turns out, the BitTorrent client is actually pleasing to use; I was foolishly expecting it to be something obtrusive, although I should have higher expectations of the HA admins. It's very transparent! What a nice implementation of distributed downloading. Plus, the author looks eerily like a friend of mine.


Okay, the original BitTorrent client was showing some upload activity before did anything to the firewall, but I opened the ports anyway.

All done and sent.

Takes about 2hrs for all samples, more if you want to be more carefull.

It took me two hours to do two samples! Some of those are really hard to ABX... I refuse to admit that 128kbps can ever be transparent!

All samples # 3 did not work on a Mac with QT 6.3 Pro

Done and e-mailed.

... maybe stylesheets?

line 19 and 20 to be more specific.

BadHorsie

I think it's because Mozilla can't (or won't) override CSS that this forum uses.

Good things come to those who wait Summer vacations are coming up, and depending on how much time is spent on design and doing it The Right Way, I refuse to believe that writing an ABC/HR app could take much more than a day to a week

That would surely be extremely welcome, Dan.

Looking forward to that.

@everyone that already sent me test results: thanks a lot.

Yeah, QT for Win chokes on them here too.

I recently added a FAAD2 OSX compile to RareWares. I suggest you try it, it will probably work.

http://pessoal.onda.com.br/rjamorim/faad_OSX.tgz

Thanks to the fella that compiled this and prefers to remain anonymous (you know who you are)

oh boy, every time you talk about QT I get fooled again.
me thinking of Qt

Hehe. Sorry.

A request: Is there any Slashdot insider here?

I would be very grateful if someone can post there about this test, and link to this page:

http://rarewares.hydrogenaudio.org/test/

Thank-you

Roberto.

Doesn't "!important" work?

Sorry to ask this so late into the test, but...



Why this, and not aacenc -streaming instead?

The tested option shows a high frequency cut off at 15505 Hz, while the -streaming cuts off at 18500 Hz. Also, isn't the general opinion to go for VBR rather than CBR?

I guess I'm showing my ignorance of most things AAC here, and perhaps the above is chosen to provide a level playing field amongst all the encoders, but aren't you looking for the best available AAC at around the 128 kbit mark? Certainly in my own testing, aacenc always gets close to 128 kbit with the -streaming profile.

Or perhaps -br 128 is the best choice with aacenc rather than -streaming quality wise?

I'm a little confused.

Den.

Indeed, VBR is best for AAC.

But using VBR profiles leads to issue of fairness. The most obvious is that CBR can't theoretically compete with VBR.

Besides, sometimes the same VBR profile can end outputting 100kbps on a sample, 150 on other. Now, comparing 150 to 128 to 100 is kinda useless. "Yeah, Psytel at 150 is better than QuickTime at 128. So what? Can it compete toe-to-toe?"

It's like when people criticized you because you were comparing Wavpack at 280 versus MPC at 190 (or something like that)

The proverbial apples vs. oranges.


now - I plan to extend this test later, comparing the best AAC encoder vs. MPC, Vorbis, MP3 and WMA. If the winner is a CBR-only codec, like SOrenson or QuickTime, we have a problem - it's unfair to compare it to MPC, that is VBR-only. So, I would end up taking this CBR winner and conduct a quick test against the highest-ranked VBR-capable encoder. And hope the VBR encodes win. >_<

But those are still just plans, and I'm working with JohnV to figure all that out.

Regards;

Roberto.

Thanks Roberto. Makes sense, level playing field and all that.

Congrats on organising the test. Good stuff!

Apples and oranges, apples and oranges... It's all coming back to me now.

Den. >_<

I'm a moron

Damn. ff123 just brought to my attention that the offsets haven't been corrected, therefore some people might have taken offset-induced artifacts as encoder artifacts (what's a flaw in the test, of course)

I take full responsability in the error, and I'm very sorry.

Some people might have to take the test again, else their results can't be usable (I'll get in touch with each one of you through e-mail. Again, very sorry). Some others seem to have ignored the offset issues and concentrated on encoder artifacs, fortunately.

New packages with corrected config files are being uploaded to the server.

For those that already downloaded the packages, you don't need to download the corrected ones - just download this file and decompress it on ABC/HR's directory, overwriting the faulty configs:
http://pessoal.onda.com.br/rjamorim/newconfigs.zip

heh, maybe I was hoping too much to have the first listening test I conduced error free. Living and learning...

Thanks a lot for your support and, once again (for the last time), my apologies.

Regards;

Roberto.

All done, results sent. I only tested one sample set without offset correction (but I didn't notice).

I would be very interested in the follow-up high quality test.

I've just come across with this: http://www.beryllium.net/~remco/linabx/

Then I remembered this test and the Linux issue, so maybe someone might find it useful. I haven't tried it myself, but looks good. It seems to be based on PCABX.

Cheers.

My results are trickling in (7 so far). How are you doing for listeners? You'll probably want at least a dozen results per sample. It would probably be much nicer if the number of listeners was double that or more.

ff123

There's a very good ABX program for Linux included with LAME.

So far, 5 people answered with full result sets, and 3 with partial results.

Also, some people said they would take their time and answer next week.

Hopefully, people will have some free time this weekend to participate in the test, at least with some of the samples.

Thanks a lot to everyone that participated so far.

Regards;

Roberto.

Yes, we do have more time for fun like this now... Just started today with two of them.

But what's wrong with the software? In the ABX window when having played one part many times I get a click at the end of the selection. It goes away if I change the slider slightly but comes back after a couple of times more. I would rate this bug as "Very annoying" on that scale.

EricS,

Can you PM me with the name of the song, and which codec number it is? I'll try to see if I can duplicate this on my machine. Also the particular starting and ending times you were using. In the ABX window, which side was the sample and which side was the reference?

If this clicking happens with every song, and with every codec, and no matter what portion you've selected, etc., just reply to this thread.

How many times do you have to play the part before it starts clicking?

ff123

It happened with both samples I have tried so far. And no matter which selection I was playing. It came after around say 30-50 switches in the abx window. And once it was there and after I changed the selection slightly to make it go away temporarily, it came back sooner - maybe 10-15 switches. I had B set to the reference. Have a TB Santa Cruz if that helps any...

I'll send you a PM if I can find out something more about it.

So you're switching between A, B, and X, and after 30 to 50 times of pressing any one of these buttons, you start to hear clicking, is that right?

Have you tried something similar using PCABX? If it happens with that program, it might not be related to ABC/HR.

I will try it out now to see what happens on my machine.

ff123

I can't verify the problem. I made a small selection of about 1 second and played it all the way through without pressing the stop button. Then I just hit A, B, or X, 50 or more times. Does this approximate what you're doing?

ff123

EricS,

One thing you might try for troubleshooting: change all the offsets to 0 (there should be two in each of the new config files which are not 0). Perhaps the clicking is related to or caused by the offset processing ABC/HR does.

Another thing to try: does the clicking happen when you switch many times in the main (abc/hr) window, or just in the abx window?

ff123

Ok. It turned out to be an error by me. So these last few messages can safely be ignored/removed...

i'm downloading the sample files right now and i will conduct the test past the next few days

btw, i'm not new to this forum, i'm just not posting anything

-andy-

Found time today. I sent you result , and some comments . I hope you can take a look to them. Anyway, thank you for this test. We need to have clear ideas on the quality reached by the publicly available aac/mp4 codec.

Bump

So when can we expect to see some results?

According to the first post, next Sunday.

I'd like to thank whoever posted about the test on slashdot.

http://developers.slashdot.org/article.pl?...tid=141&tid=188

On behalf of Dibrom, or the community?

Slashdottings are not fun.

In behalf of the test, of course. Test quality will increase if more people participate in it.

Just a few of the gems so far in the /. comments

http://developers.slashdot.org/comments.pl...031&cid=6157272
http://developers.slashdot.org/comments.pl...031&cid=6157415
http://developers.slashdot.org/comments.pl...031&cid=6157454
http://developers.slashdot.org/comments.pl...031&cid=6157462

Great ones. But I think the winner is link #2