Stuff for the AAC 96 kbps listening test, Bitrate tables, sample files, etc Options

[Alex B]

A bitrate table in Excel format (zipped).

 Bitrates_for_AAC_96kbps_listening_test.xls.zip ( 5.13K ) Number of downloads: 174

The related post: http://www.hydrogenaudio.org/forums/index....st&p=762128

[IgorC]

For the same topic.

Attached File(s)

 bitrate_per_member.zip ( 2.19K ) Number of downloads: 69

 

[/mnt]

Another bitrate table.

EDIT: Uploaded updated table

This post has been edited by /mnt: Jul 6 2011, 23:47

Attached File(s)

 96kbps_AAC_Bitrates__mnt_.zip ( 3.35K ) Number of downloads: 59

 

[/mnt]

At 96kpbs - 100kbps Nero, FhG and QT has trouble with this sample.

Attached File(s)

 Jerkin__Back__n__Forth.flac ( 3.04MB ) Number of downloads: 234

 

[hellokeith]

Dream Theater track "These Walls" from the album Octavarium, sample range is 2:34 - 3:04.

Nero -q 0.345 - The cymbal stereo imaging gets messy, not too difficult to ABX with headphones:

Attached File(s)

 Dream_Theater___These_Walls.wv ( 3.46MB ) Number of downloads: 89

 

[Steve Forte Rio]

High frequency and total detalization loss for all AAC encoders under test (except FhG - it gives very high bitrate, so there is a little High Frequencies lost).

Attached File(s)

 Bach___Suite_no.1_Allemande__harpischord_.flac ( 2.98MB ) Number of downloads: 98

 

[/mnt]

A yet another sharp transient problem sample.

However this sample gives each encoder it's characteristics. For example Nero does terrible with the low freq transients while sounding fine with the high freq one; Quick Time does ok with the low freq transients, but tremolo disortion appears during or after the high freq transients appear.

Attached File(s)

 Headhunter.flac ( 3.46MB ) Number of downloads: 101

 

[muaddib]

The previous posts show why samples had to be chosen long time before even thinking of the listening test organization.
I hope that at least some random process for choosing final samples will be involved. For example using some public lottery as I suggested long time ago:
http://www.hydrogenaudio.org/forums/index....showtopic=64991

This post has been edited by muaddib: Jul 8 2011, 07:57

[IgorC]

It's interesting idea.
A simple solution will be online dice-roller.

http://rolz.org/?p=group#

/mnt and Steve Forte Rio probably could make a list with 100-200 samples. Then we all (interested in this test) will enter into chat with dice-roler function and will start to roll the dice.

What do you think?
Seems very good to me. Hell, I like it.

This post has been edited by IgorC: Jul 8 2011, 10:23

[hellokeith]

The previous posts show why samples had to be chosen long time before even thinking of the listening test organization.

How many samples at minimum do we think the test will need?

[IgorC]

20

[Sebastian Mares]

20 samples is quite a lot depending on the quality of the encoders.

[IgorC]

There were 30 samples in previous test. Large number of samples helps to get more statistically valid results.
It was a good decision (not a coincidence).
The only price to pay is some extra time for recollecting the samples.

So, there will be 20 samples in this test.



As of random choice of samples.
/mnt and Steve Forte Rio will prepare the list with large number of samples (from previous tests, upload section, samples of other members etc..) and the random choice will be performed online here http://rolz.org/?p=group#

We will define the time of this event shortly.
All members and developers can participate.

People, please, submit your samples (untill tuesday).Also it will be better to indicate the interval.
For example the sample can have 30 seconds of the lenght but more interesting part is between 14-22 seconds.

[Steve Forte Rio]

Well, here is some samples from my collection. Most of them causes pre-echo, the last sample is for stereo image testing (better listen through the loudspeakers).

tiesto-athena.flac (13-28s)
tiesto-do_you_feel_me.flac (1-21s)
tydi-meet_me_in_kyoto.flac (first 15s)
(genre: electronica)

above_and_beyond-on_a_good_day.flac (first 15s)
(genre: electronica/vocal trance)

And more:

Sound Expert nine samples:

bachjs.flac
bass.flac
castanets.flac
frmalspeech.flac
glockenspiel.flac
harpsichord.flac
lofi.flac
moldfield.flac
quartet.flac

This post has been edited by Steve Forte Rio: Jul 9 2011, 10:32

[IgorC]

Great, SFR.

It will be random choice so it's ok to include a pile of samples from

http://rarewares.org/test_samples/
http://ff123.net/samples.html
Roberto's and Sebastian's tests http://listeningtests.t35.com/
http://membres.multimania.fr/guruboolez/AUDIO/
Upload section http://www.hydrogenaudio.org/forums/index.php?showforum=35
http://www.mp3-tech.org/tests/aac_48/samples/

This post has been edited by IgorC: Jul 9 2011, 11:44

[lvqcl]

The quality of ffmpeg @128kbps isn't low enough, IMHO.

[IgorC]

What settings do you use?
http://www.hydrogenaudio.org/forums/index....st&p=761300
ffmpeg -i file -acodec aac -ab 128k -strict experimental file.m4a

if it will be necesarry we can lower the bitrate to 96 kbps if the quality will be too high for low anchor.

This time low anchor shouldn't be that bad as in previous test. (I've tried ffmpeg's AAC 128 kbps on several samples and it was very bad)

[/mnt]

These 2 samples don't do too well with AAC at low bitrates.

Does anyone have a list and links to the samples that was used on the recent 64kbps listening test?

[Steve Forte Rio]

Another one sample from my music library. Causes high freq losses and signal correlated noise in some cases.

Aria_Finch_han_dal_vino.flac
(Symphonic orchestra + male vocal)

2lvqcl
Try to encode this sample with ffmpeg (parameters given above) and you will hear that it's encoding quality is low enough

It will be random choice so it's ok to include a pile of samples from

http://rarewares.org/test_samples/
http://ff123.net/samples.html
Roberto's and Sebastian's tests http://listeningtests.t35.com/
http://membres.multimania.fr/guruboolez/AUDIO/
Upload section http://www.hydrogenaudio.org/forums/index.php?showforum=35
http://www.mp3-tech.org/tests/aac_48/samples/

Thank you! I'm going to post a large list of samples from this locations tomorrow (hopefully, I'll get the direct links). In order to have them all at one place, here

This post has been edited by Steve Forte Rio: Jul 9 2011, 21:09

[IgorC]

Does anyone have a list and links to the samples that was used on the recent 64kbps listening test?

First 15 samples of them are from previous public test (2007)
http://www.hydrogenaudio.org/forums/index....st&p=749072

http://listening-tests.hydrogenaudio.org/i...all_samples.zip

1.poets of the fall
2. bachpsichord
3.spmg54_1 (german speech)
4. tomsdiner
5. kraftwerk
6.macabre
7. bibilolo
8.Stravinskii Capriccio
9. fly_to_fly
10. fatboy
11. les voleurs
12. eig
13. enter sandman
14. big yellow
15. davinci
16. Castanets
17. Linchpin
18. waiting
19. Bittersweet
20. Dimmu borgir orchestra
21. bleed
22. Sangue de bairro
23. Against
24. stars
25. worlds apart
26. take your finger frin my head
27. i want to break free
28. atrain
29. bonhemian rhapsody
30. on the roof with Quasimodo

[IgorC]

If the samples from the last test were mentioned then it's actually fair to include these samples as well.
Anyway it will be a random choice on a big amount of samples.

It shouldn't be an issue IMO.

This post has been edited by IgorC: Jul 10 2011, 00:41

[hellokeith]

Anyway it will be a random choice on a big amount of samples.

Purely random? What if the majority of the samples end up being marimbas and guys yodeling in the shower?

I suggest categories of samples, with random selection from each category.

[IgorC]

I suggest categories of samples, with random selection from each category.

Well, it's up to /mnt and Steve Forte Rio. They will control the random selection of the samples. Maybe if they are kind to do some extra work it will be great.

This post has been edited by IgorC: Jul 10 2011, 02:41

[Steve Forte Rio]

Hm... What categories do you mean? By genres or maybe by the type of artefacts?


I've just started to fill the table with sample names, URL's, genres (only for samples posted here) and comments (if exist). Anyway, I think it will take very long time to verify every sample (of more than 200 items) and determine the genre or artefacst type for each of them.

What do you think, /mnt? Also, does anyone have some suggestions?

[IgorC]

What if the majority of the samples end up being marimbas and guys yodeling in the shower?

There won't be such samples at all.

The big list will contain a lot of samples with different kind of possible artifacts (temporal, frequency, stereo and other distortions) .
Probability will work by itself. I'm pretty sure we will have the same results with/without classifying by artifacts.

I suggest categories of samples, with random selection from each category.

Every suggestion has a down side.
It will be very difficult to classify artifacts. Somebody hears some temporal distortion other hears stereo issues. The artifacts are rarely have a single nature.
We will have days and days of hot discussions and probably will end with the same situation as purely random choice.


Everybody is welcome to online random selection of the samples. Here http://rolz.org/?p=group#
This Wednesday at 12:00 (GMT - 3)