...then we can use your help!

We are working on the low bitrate AAC encoders, and need some more feedback to make decisions.

At low bitrates, sometimes what sounds best is an issue of what sounds least bad...so we want to know what you think sounds least bad At the same time you can get an impression what is up-and-coming in AAC.

Download this first:

http://www.audiocoding.com/listening_tests/decode.rar

Links to samples:

http://www.audiocoding.com/listening_tests/sample1.rar
http://www.audiocoding.com/listening_tests/sample2.rar
http://www.audiocoding.com/listening_tests/sample3.rar
http://www.audiocoding.com/listening_tests/sample4.rar
http://www.audiocoding.com/listening_tests/sample5.rar
http://www.audiocoding.com/listening_tests/sample6.rar
http://www.audiocoding.com/listening_tests/sample7.rar
http://www.audiocoding.com/listening_tests/sample8.rar
http://www.audiocoding.com/listening_tests/sample9.rar
http://www.audiocoding.com/listening_tests/sample10.rar
http://www.audiocoding.com/listening_tests/sample11.rar

Put them all in the same dir and run the decode.bat.

And compare the A B and C files to the reference. Grade them from
0 to 5 and make a few notes on what your impressisons of them are.

ABC/HR may come in handy for that:
http://ff123.net/abchr/abchr.html

Send your results to gcp@sjeng.org. Thanks for the help!

When will this test end? I would be glad to participate, but I can only start on Saturday.

Windows only?

These lines in decode.bat don't seem to be correct.

Was just about to comment on that... it doesn't only "seem" to not be correct, it's obviously not should be easy enough to fix for most people, before garf fixes the .bat himself

Sorry for that - file is fixed now

Should be possible on Linux too, though you'll have to decode the files manually. The public FAAD source can handle them. So basically you just have to find an AAC and FLAC decoder for Linux or your OS of choice...

Thanks to everyone that sent results so far!

One more question: will the OS X binary from rarewares work for decoding these files?

http://www.rarewares.org/aac.html

THIS i can do, but, as others have asked, when will it end? i get back from Cali on friday...

It is dated 2003-09-15, which would be too old...we fixed and improved a lot of things since then.

Maybe someone can do a new compile based on new FAAD sources.

Yeah, we have about a week time.

is this a parametric stereo (+sbr) decoder or does it include more goodies?

FAAD2 decodes SBR and PS. It has done so for a while now...?! Both are needed to decode the files in this test.

Ok, fast results are sent.
EDIT: BTW, rather impressive quality for such a bitrates...

Now I have some spare time to analyze what I've done
So, I have some questions now:
1. What codecs/settings were used for this test ?
(If this is not a secret period)
2. (Not directly related to this test) Are the spatial audio specs finalized ? And is there any chance that it will be used by AHEAD aac encoder ?
Thanks in advance !!

Combinations of SBR and PS, all from our in-development encoder.



They are not finalized. Once they are I assume we will do something in this area.

A very big thanks from the NeroDigital team to everyone that submitted results!

Can you tell us something more about the results?

We now know the break-even point for current Parametric Stereo.

Hey Garf, in that sample1 of the pack we downloaded I did some goofing around. I took the flac samp and re-encoded at 80kbits MP3Pro and 80kbits HEAAC and the MP3Pro samp sounded much better. There was alot of ringing in the begining that wasn't present or pretty much unaudible in the MP3Pro re-encode(also at 96kbits MP3Pro still sounded better at 80kbits).

Just something you guys may want to check out, since you are competing with that codec.


***Edited Part and Note***
The gain/volume on the mp3Pro samp was slightly less than the original on that triangle sounding instrument. So there was audible difference, but the ringing wasn't present.

Hehe
Can you be a bit more specific, if you have spare time, of course ?
Just curious about settings used for all 3 encodes and overall conclusion

Well - the overall conclusion is that PS48 and SBR48 are pretty much equal:

HE64: 3.61
HE48: 2.78
PS48: 2.75

At current state, PS and HE were very close - where HE failed, PS was ranked as very good - and vice versa.

Current conclusion is not to use PS on this bit rate - due to power consumption and backwards compatibility - at 40 Kb/s, PS brings significant improvements, and will be used.

Thanks for clarification, Ivan !
So, As I understand this, now situation with nero aac implementation will be:
(for stereo)
>96 -> LC
48-96 -> HE
<48 -> PS
So, we all waiting for the new Nero version to test it by ourselves !
P.S. It would be great, if PS will be finised until new Roberto test begins.
I hope, it will happen...

Thanks again.

There's also DSBR which can be useful from about 64kbps to 128kbps. Finding the exact margins for optimal quality also needs a lot of testing there.

Just in case you were wondering why we're taking so long

I remeber, Ivan mentioned somewhere that your testing team try DSBR to find if it is better on this bitrates...
Is this clarified now, that downsampled SBR produce better result using 128Kbit bitrate, than pure LC ?

Understandable. But life is short, so we need it all and imidiately
EDIT: grammar

Hi Garf and Ivan Dimkovic. Its cool to see the two of you doing development for Ahead AAC and to see that your efforts and talents being put to very good use for the masses!

Do you have any timeframe for an updated AAC encoder?

We want to participate in roberto's test with the new encoder so there should be an update before it starts.

It won't contain all the new stuff though, need some more time for that.

Good luck because it is planned for the end of the month...

Great, thanks for the fast reply. I'm about ready to try some 64-96 kbps Nero AAC out myself (for use in a mobil phone), so any upcoming improvement is most welcome!

Well, it's another one of those pesky situations where we're near the break-even point. Because of the nice feedback we got with this test (which is VERY MUCH APPRECIATED), we may do a public test of 128k DSBR too.

Do you have mobile phone which can play AAC-HE?

Errr, probably not. I didn't read up on the HE-AAC issue before posting.

My brandnew SE K700 can play the HE-AAC test files I tried on it, but that of course doesn't prove anything. I seem to remember something about HE-AAC files just being played as mono if the player doesn't support HE? Garf?

PS AAC is played as mono if the player doesn't support it.

HE AAC is played as LC AAC if the player doesn't support it.

Thanks for the input, Garf. This is what the SE K700 whitepaper say:



I guess I will try and see whether CBR80 LC or maybe CBR96 LC will do decent enough. It is just for casual use on a mobile phone after all, so unless there is too many artifacts, those may make a very decent trade off.

DSBR will greatly enhance quality for non-HE AAC capable players that have to play back HE-AAC content.

Current HE-AAC is not very efficient at that bitrate yet, so if you have just a little more patience you will be in for some pleasant surprises

Sounds delightful!

(I guess I did good by buying Nero 6 after all then)