Skip to main content

Topic: Spoken word sample easily ABX-able at Nero Q .50 (Read 6350 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • matt_t
  • [*]
Spoken word sample easily ABX-able at Nero Q .50
This seems to be some kind of stereo problem. There are two readers, one slightly right of centre, the other slightly left. But there is some "leakage" - by no means offensive, but by no means transparent either.

I converted to Nero AAC using dBpoweramp, forced LC (en passant - why is it "always" advised to use "no forced profile" when many players will not play HE and so on? If your player requires LC, why not force it? Just asking!), quality .50.

foo_abx 1.3.3 report
foobar2000 v0.9.5.6
2008/11/27 23:09:14

File A: C:\Documents and Settings\Matthew\My Documents\My Music\Tests\CD1-02.flac
File B: C:\Documents and Settings\Matthew\My Documents\My Music\Tests\CD1-02.m4a

23:09:14 : Test started.
23:10:40 : 01/01  50.0%
23:11:25 : 02/02  25.0%
23:12:38 : 03/03  12.5%
23:13:21 : 04/04  6.3%
23:13:51 : 05/05  3.1%
23:14:42 : 06/06  1.6%
23:15:16 : 07/07  0.8%
23:16:13 : 08/08  0.4%
23:17:01 : 09/09  0.2%
23:17:55 : 10/10  0.1%
23:18:35 : 11/11  0.0%
23:19:39 : 12/12  0.0%
23:20:39 : 13/13  0.0%
23:21:18 : 14/14  0.0%
23:21:57 : 15/15  0.0%
23:22:38 : 16/16  0.0%
23:23:12 : 17/17  0.0%
23:23:17 : Test finished.

----------
Total: 17/17 (0.0%)

I'll put the FLAC in the uploads forum. That seems to be the logical place for it. Apologies if this is not SOP, but I'm a relative noob.

  • muaddib
  • [*][*][*][*]
  • Developer
Spoken word sample easily ABX-able at Nero Q .50
Reply #1
Of course that you should force LC when you want to play a file on a device that doesn't support HE.
However LC is default for q 0.50.
It would be good if somebody else could confirm the problem.
Thank you for reporting this one. We'll have it in mind.

Spoken word sample easily ABX-able at Nero Q .50
Reply #2
You don't always have to force LC with the Nero AAC encoder as Nero will default to LC when using certain -q values.  You only have to force LC when using a -q value that produces lower bitrates.

What overall average bitrate did the track end up having with Nero at -q0.5?  Also, would you mind posting a FLAC/WAV 30 second sample so that we can encode the track and conduct ABX tests for ourselves?

  • muaddib
  • [*][*][*][*]
  • Developer
Spoken word sample easily ABX-able at Nero Q .50
Reply #3
The sample is here:

http://www.hydrogenaudio.org/forums/index....showtopic=67593

What is missing is the version of Nero AAC Encoder used.

  • matt_t
  • [*]
Spoken word sample easily ABX-able at Nero Q .50
Reply #4
The sample is here:

http://www.hydrogenaudio.org/forums/index....showtopic=67593

What is missing is the version of Nero AAC Encoder used.


Sorry, would forget my head if it wasn't screwed on! It was the latest version, 1.3.3.0.

  • [JAZ]
  • [*][*][*][*][*]
Spoken word sample easily ABX-able at Nero Q .50
Reply #5
23:10:40 : 01/01  50.0%
23:23:12 : 17/17  0.0%


Ok, you definitely ABXed it.  But taking 13 minutes for 17 tries is quite an indicative that it was not easy at all to hear the difference.

I'm adding this, just in case. Of course, improvements are always a desiderable thing for codecs.

  • matt_t
  • [*]
Spoken word sample easily ABX-able at Nero Q .50
Reply #6
Quote
' date='Nov 28 2008, 17:34' post='601825']


23:10:40 : 01/01  50.0%
23:23:12 : 17/17  0.0%


Ok, you definitely ABXed it.  But taking 13 minutes for 17 tries is quite an indicative that it was not easy at all to hear the difference.

I'm adding this, just in case. Of course, improvements are always a desiderable thing for codecs.


Jaz, I totally disagree, for the following reasons:

1) You weren't watching over my shoulder as I took the test. As it happens, since 16 tests is considered a decent sized sample, I made a note after each test to keep count, and then added one for luck, just to keep it shipshape.

2) Listening to the same few seconds repeatedly is enough to fry anyone's brain, so I tend to have a few seconds' rest.

3) I make sure I am going to press the right button, so I ask myself whether I really mean A=X or B=X rather than risk clicking the wrong button by mistake. I am a pretty crazy mouse user, always clicking something I didn't mean to!

4) In any case, as I understand it, ABX is not a speed or endurance test.

From this thread: http://www.hydrogenaudio.org/forums/index....showtopic=16295

"Corollary : only give answers of which you are absolutely certain ! If you have the slightest doubt, don't answer anything. Take your time. Make pauses. You can stop the test and go on another day, but never try to guess by "intuition". "

5) With all this in mind, and even without it, I'd say that less than 1 minute each for 17 answers is bordering on greased lightning!


Tangentially related info: I have found it extremely difficult, bordering on impossible, to ABX harpsichord music at Q .50 - .55, whereas I can even at LAME V1. Bravo to Nero for that, since I have loads of it in my collection. Anyone for JS Bach...?  So I was surprised that a spoken word sample foxed it.

  • [JAZ]
  • [*][*][*][*][*]
Spoken word sample easily ABX-able at Nero Q .50
Reply #7
I shouldn't have drawn that conclusion so fast. Apologies for it.

I've done an ABX too, to verify what was the problem exactly:

Range ABX'ed :  00:01.5 - 00:02.9, but with the woman voice it was also noticeable.

foo_abx 1.3.3 report
foobar2000 v0.9.5.6
2008/11/29 11:25:29

File A: C:\Users\JosepMa\Downloads\CD1_02.mp4 ( -q .50 -> 131kbps )
File B: C:\Users\JosepMa\Downloads\CD1_02.wav

11:25:29 : Test started.
11:28:31 : Trial reset.
11:31:21 : 01/01  50.0%
11:31:26 : 02/02  25.0%
11:31:45 : 03/03  12.5%
11:31:55 : 04/04  6.3%
11:32:01 : 05/05  3.1%
11:32:13 : 06/06  1.6%
11:32:32 : 07/07  0.8%
11:32:39 : 08/08  0.4%
11:32:51 : 09/09  0.2%
11:33:01 : 10/10  0.1%
11:33:09 : 11/11  0.0%
11:33:13 : 12/12  0.0%
11:33:19 : 13/13  0.0%
11:33:23 : 14/14  0.0%
11:33:31 : 15/15  0.0%
11:33:36 : 16/16  0.0%
11:33:55 : Test finished.

----------
Total: 16/16 (0.0%)


The problem, as you describe, is heard as a more centered voice than in the original, plus also a slight reverb or less clean voice than the original.

I'm not sure if i would hear this without an ABX, though.

  • matt_t
  • [*]
Spoken word sample easily ABX-able at Nero Q .50
Reply #8
No worries Jaz.

Anyhow, I just tried it at Q .55, and guess what:

foo_abx 1.3.3 report
foobar2000 v0.9.5.6
2008/11/29 22:10:52

File A: C:\Documents and Settings\Matthew\My Documents\My Music\Tests\CD1-02.flac
File B: C:\Documents and Settings\Matthew\My Documents\My Music\Tests\CD1-02.m4a

22:10:52 : Test started.
22:13:02 : 01/01  50.0%
22:14:05 : 02/02  25.0%
22:15:16 : 03/03  12.5%
22:16:29 : 04/04  6.3%
22:17:26 : 05/05  3.1%
22:18:52 : 06/06  1.6%
22:19:52 : 07/07  0.8%
22:20:51 : 08/08  0.4%
22:22:47 : 09/09  0.2%
22:25:17 : 10/10  0.1%
22:27:10 : 11/11  0.0%
22:28:54 : 12/12  0.0%
22:30:16 : 13/13  0.0%
22:31:50 : 14/14  0.0%
22:33:17 : 15/15  0.0%
22:34:43 : 16/16  0.0%
22:34:50 : Test finished.

----------
Total: 16/16 (0.0%)


Maybe I'll see if I can stand doing another session at Q .60. Not yet, though!

  • matt_t
  • [*]
Spoken word sample easily ABX-able at Nero Q .50
Reply #9
I haven't done a proper ABX test, but it's immediately clear (to me at least) that the new 1.5.1.0 encoder is MUCH better on this sample. I recall it being easy to ABX even at q.55 with 1.3.3.0, whereas I can imagine q.4 being hard with 1.5.1.0.

So congratulations to the Nero developers!