Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Request for a quick listening test (Read 14217 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Request for a quick listening test

Hi,

We are testing couple of coding modes for 32, 40, 48 and 64 kb/s bit-rates and similar VBR presets, and I would like to get community feedback on couple of possible modes:

Test archive contains 10 samples, encoded with various encoding modes at

24, 32, 40, 48 and 64 kb/s  - LC AAC @64 is used as "anchor"

Test archive is 27 MB:
http://www.audiocoding.com/bigtest.rar

Codec list (encrypted - password will be provided in 15 days):
http://www.audiocoding.com/codecs.rar

Just download the bigtest.rar, unpack it to a directory - and click on 'decode.bat'

Please use Java ABC/HR with the config files as described:

http://www.hydrogenaudio.org/forums/index....ndpost&p=192426


Send the results to this address:  idimkovic (AT) nero DOT com

Thanks in advance!

Request for a quick listening test

Reply #1 –
First of all, thanks to Ivan for giving us the opportunity to have a glance at the latest development in AAC.

For convenience, I've made some config files, which you can use with ff123's ABC/HR app or with ABC/HR for Java. Just unzip them in the same folder as the decode.bat. These config files contain offset and gain correction, which is quite significant with some samples (offsets are corrected by up to 35ms, gain up to 2dB).

Request for a quick listening test

Reply #2 –
Schnofler,

Thanks a lot for your work!

Request for a quick listening test

Reply #3 –
Quote
First of all, thanks to Ivan for giving us the opportunity to have a glance at the latest development in AAC.

For convenience, I've made some config files, which you can use with ff123's ABC/HR app or with ABC/HR for Java. Just unzip them in the same folder as the decode.bat. These config files contain offset and gain correction, which is quite significant with some samples (offsets are corrected by up to 35ms, gain up to 2dB).

Please note that the native windows abc/hr cannot correct for gain, only for offsets.

ff123

Edit:  I wonder why there would be such a big difference in gain?

Request for a quick listening test

Reply #4 –
Quote
Edit: I wonder why there would be such a big difference in gain?

Yes, I was surprised, too. I checked the values with Replaygain, and it produced pretty similar results. On some samples this is very noticeable (compare the reference with codec C on sample 7, for example; the reference is clearly louder, replaygaining yields track gains of -2.5dB vs. +0.8dB).

Quote
Schnofler,

Thanks a lot for your work!

No problem, ABC/HR-Java does this almost automatically.

Request for a quick listening test

Reply #5 –
Quote
Edit: I wonder why there would be such a big difference in gain?


I will explain at the end of the test

Request for a quick listening test

Reply #6 –
Quote
Quote
Edit: I wonder why there would be such a big difference in gain?

Yes, I was surprised, too. I checked the values with Replaygain, and it produced pretty similar results. On some samples this is very noticeable (compare the reference with codec C on sample 7, for example; the reference is clearly louder, replaygaining yields track gains of -2.5dB vs. +0.8dB).

Quote
Schnofler,

Thanks a lot for your work!

No problem, ABC/HR-Java does this almost automatically. 

If there are tracks which are clearly louder, it is probably best to stick with abchr-java.  Typically louder sounds better, all else being equal.

ff123

Request for a quick listening test

Reply #7 –
True - please use Java version for the test, as some new algos do have these issues >at the moment<

Request for a quick listening test

Reply #8 –
Argh.. Looks like I messed up again. ABC/HR-Java doesn't read the gain values correctly from config txt-files (xml and encrypted configs work fine).

There are two ways to correct this: use the xml configs from the zip at the end of this post, or get a new ABC/HR-Java version here. This is a development version, and I didn't have time to test it properly, so there might be some bugs. For those who encountered program freezes in Roberto's AAC test, this version might help (reports are welcome, although I will open a thread about this when I release a new proper version).

Request for a quick listening test

Reply #9 –
This is not related directly to the request.  It's about the included aacdec.exe app.

channel configuration -- all the monos get made into stereo
when going from adif to adts using

  aacdec -a sample1_A_aacdec.aac  sample1_codecA.aac

\tmp\aac_tmp>comp s1_cA_faad.aac    sample1_A_aacdec.aac
        Comparing s1_cA_faad.aac and sample1_A_aacdec.aac...
Compare error at OFFSET 3
file1 = 40
file2 = 80
Compare error at OFFSET 21
file1 = 40
file2 = 80
Compare error at OFFSET 39
file1 = 40
file2 = 80
Compare error at OFFSET 1C8
file1 = 40
file2 = 80
Compare error at OFFSET 278
file1 = 40
file2 = 80
Compare error at OFFSET 32F
file1 = 40
file2 = 80
Compare error at OFFSET 3CE
file1 = 40
file2 = 80
Compare error at OFFSET 489
file1 = 40
file2 = 80
Compare error at OFFSET 533
file1 = 40
file2 = 80
Compare error at OFFSET 5DD
file1 = 40
file2 = 80
10 mismatches - ending compare

faad used:
*********** Ahead Software MPEG-4 AAC Decoder V2.0 RC3
Build: Feb  5 2004

aacdec used (included in that package):
*********** Ahead Software MPEG-4 AAC Decoder V2.1 beta
Build: Mar  1 2004

Request for a quick listening test

Reply #10 –
Quote
channel configuration -- all the monos get made into stereo

Yup... good observation 
Quote
when going from adif to adts using

  aacdec -a sample1_A_aacdec.aac  sample1_codecA.aac

Hmm, the -a switch might be broken on these files. I'll have to check.

Menno

Request for a quick listening test

Reply #11 –
Thanks for the config files, schnofler. They're very helpful.

<bugreport>
The program crashes on me from time to time, though, so I have to save my sessions midway, because I can rarely complete a session without a crash (or two). This is ABC/HR v0.4b3 SE, 16 Jan. I'll play a sample, and the sample will cut out. At this point, I can save my session, close the window, open ABC/HR again, and load the session, and continue playing all the samples. If I press play a second time, though, the window stops responding, and I have to kill it with Task Manager. This is on a dual-CPU machine running Windows XP Pro, SP1, running Sun's VM version 1.4.2_03-b02. Soundcard is TBSC with 4193 drivers. What else can I tell you? It doesn't happen immediately, only after playing a few dozen times. It happened on almost every clip, regardless of the time slider position. It never happened when I played the reference, but then again, 90% of my play time was with the compressed versions.

EDIT: problem resolved with ABC/HR v0.4b4.
</bugreport>

I have a question about my sample rating procedure:

On one sample, a certain artifact (such as stereo collapse) makes a big difference in the overall impression of the sample, and on another, the artifact doesn't hurt the sample's overall feel much. Although the artifact is clearly audible in both cases, I mark the affected test clip way down on the first sample and only a little on the second. Is this the proper way to rate samples, by how pleasant they are to listen to (in comparison to the reference), or should I be going strictly by the audibility of artifacts? Basically, I'm weighting the importance of each artifact differently for each sample.

Request for a quick listening test

Reply #12 –
PS?

Request for a quick listening test

Reply #13 –
Quote
PS?

Definitely

Request for a quick listening test

Reply #15 –
Erm... PS = Parametric Stereo. The new low bitrate enhancement tool being standardized into MPEG4 AAC.

Request for a quick listening test

Reply #16 –
...and beeing incorporated into FAAD2

Request for a quick listening test

Reply #17 –
Already done  Check the CVS.

Menno

Request for a quick listening test

Reply #18 –
Results sent!

Request for a quick listening test

Reply #19 –
Wanted to make sure:
this means FAAD2 derived filters will definitely play back HE-AAC+PS,
as in CoreAAC will handle it with the benefits

I can't wait! (well i can a little)

Request for a quick listening test

Reply #20 –
Quote
I have a question about my sample rating procedure:

On one sample, a certain artifact (such as stereo collapse) makes a big difference in the overall impression of the sample, and on another, the artifact doesn't hurt the sample's overall feel much. Although the artifact is clearly audible in both cases, I mark the affected test clip way down on the first sample and only a little on the second. Is this the proper way to rate samples, by how pleasant they are to listen to (in comparison to the reference), or should I be going strictly by the audibility of artifacts? Basically, I'm weighting the importance of each artifact differently for each sample.

I think so - after all the grading scale is how annoying it sounds, not how many artifacts you can identify.

Request for a quick listening test

Reply #21 –
Quote
I think so - after all the grading scale is how annoying it sounds, not how many artifacts you can identify.

Okay, thanks for the feedback, Garf. I wasn't sure whether to grade purely on the sample's closeness to the original ("this sample sounds closest to the original, except the hi-hat artifact is so annoying!"), or to allow the "enjoyment factor" to affect my grading as well ("it sounds a bit lo-fi, but I'd prefer that my encodings resemble this sample more than the other samples").

Request for a quick listening test

Reply #22 –
woohoo! i'm digging 2, if not 3 of the samples...  have these *always* been the samples? (or at least as of late? ..late being the last 2-3 tests)

edit: all done. that was a really fun test

Request for a quick listening test

Reply #23 –
Some of them were in the other tests, most of these seem new ones from Ivan.

Request for a quick listening test

Reply #24 –
this should be interesting. i like doing 1CD rips, but for longer movies, i have to use AAC-HE, which unfortunately destroys a lot of the surround type info, IIRC (from my personal experiences). i could hear in some of these tests with just a 2 speaker (headphone) setup how some preserved the sound field. i guess that's why it's called Parametric Stereo