-qo in 3.96

Topic: -qo in 3.96 (Read 11301 times) previous topic - next topic

0 Members and 2 Guests are viewing this topic.

-qo in 3.96

2004-08-02 17:04:02

Uh right here we go.

(Clip from Queen's "Bicycle Race" ripped from The Platinum Collection (Disc 1))

-qo in 3.96

Reply #1 – 2004-08-02 17:13:45

Put a link to the thread the sample refers to.

-qo in 3.96

Reply #2 – 2004-08-02 17:29:24

SAMPLE: biniou.wav (I remember that I've uploaded this sample long time ago; I don't know if it's still available)
EDIT: link is http://membres.lycos.fr/guruboolez/AUDIO/s...r_ahead_aac.piz (rename .piz to .zip)
RANGE: 0:05.5' - 0:08.0'
SETTINGS: 3.96.1 --preset cbr 128 vs 3.96.1 --preset cbr 128 -q0

Code: [Select]

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/02 17:57:55

File A: file://C:\biniou CBR128.mp3
File B: file://C:\biniou CBR128 Q0.mp3

17:57:58 : Test started.
17:59:49 : 00/01  100.0%
17:59:57 : 00/02  100.0%
18:00:02 : 01/03  87.5%
18:00:13 : 01/04  93.8%
18:00:20 : 02/05  81.3%
18:00:30 : 03/06  65.6%
18:00:34 : 04/07  50.0%
18:00:39 : 05/08  36.3%
18:00:43 : 06/09  25.4%
18:00:48 : 07/10  17.2%
18:00:53 : 08/11  11.3%
18:00:58 : 09/12  7.3%
18:01:03 : 10/13  4.6%
18:01:13 : 11/14  2.9%
18:01:22 : 12/15  1.8%
18:01:27 : 13/16  1.1%
18:01:32 : 14/17  0.6%
18:01:36 : 15/18  0.4%
18:01:46 : 16/19  0.2%
18:01:59 : 17/20  0.1%
18:02:03 : Test finished.

 ---------- 
Total: 17/20 (0.1%)

The first file (cbr by default) is less distorted. -q0 switch add some audible ringing. Two comments:
• the 4 first trials were performed on another testing range.
• the ringing could be visualize in a frequency editor: there's a lot of missing information between the two encodings, and -q0 looks like a Gruyère cheese. I suppose that developers could work on the problem without listening test.
• 3.97a3 at -q0 is identical to 3.96.1 at the same setting.

SAMPLE: Dom Bedos.wav - organ, available on ftp://ftp2.foobar2000.net/foobar/Dom_Bedos.zip
RANGE: four first seconds
SETTINGS: 3.96.1 --preset cbr 128 vs 3.96.1 --preset cbr 128 -q0

Code: [Select]

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/02 18:21:39

File A: file://C:\temp\HA PACK\__samples\Dom Bedos#3 CBR128.mp3
File B: file://C:\temp\HA PACK\__samples\Dom Bedos#3.mp3

18:21:42 : Test started.
18:21:52 : 01/01  50.0%
18:22:01 : 02/02  25.0%
18:22:04 : 03/03  12.5%
18:22:08 : 04/04  6.3%
18:22:11 : 05/05  3.1%
18:22:15 : 06/06  1.6%
18:22:20 : 07/07  0.8%
18:22:23 : 08/08  0.4%
18:22:26 : 09/09  0.2%
18:22:30 : 10/10  0.1%
18:22:34 : 11/11  0.0%
18:22:38 : 12/12  0.0%
18:22:41 : 13/13  0.0%
18:22:44 : 14/14  0.0%
18:22:47 : 15/15  0.0%
18:22:51 : 16/16  0.0%
18:22:53 : Test finished.

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

Here, the ringing introduced by -q0 is simply awfull!! (and of course, obvious). I suppose that even untrained people could hear the problem. Again, the difference and the problem with -q0 could be visualized on a frequency editor.

I think that NeoRenagade had just founded a real problem with lame -q0.

EDIT: link for biniou sample added

-qo in 3.96

Reply #3 – 2004-08-02 17:50:01

SAMPLE: Platée I,6.wav - orchestral
LOCATION: ftp://ftp2.foobar2000.net/foobar/Platee.zip
RANGE: full sample, but beginning was enough
SETTINGS: 3.96.1 --preset cbr 128 vs 3.96.1 --preset cbr 128 -q0

The two previous samples were highly tonal (organ and biniou, cousin of bagpipes); here is something more 'noisy' and less linear: a baroque orchestra, playin with fury, with harpsichord in background.

Code: [Select]

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/02 18:40:33

File A: file://C:\temp\HA PACK\__samples\Platée I,6 CBR128.mp3
File B: file://C:\temp\HA PACK\__samples\Platée I,6 CBR128 Q0.mp3

18:40:34 : Test started.
18:40:46 : 01/01  50.0%
18:40:49 : 02/02  25.0%
18:40:53 : 03/03  12.5%
18:40:57 : 04/04  6.3%
18:41:00 : 05/05  3.1%
18:41:07 : 06/06  1.6%
18:41:11 : 07/07  0.8%
18:41:15 : 08/08  0.4%
18:41:18 : 09/09  0.2%
18:41:21 : 10/10  0.1%
18:41:25 : 11/11  0.0%
18:41:28 : 12/12  0.0%
18:41:31 : 13/13  0.0%
18:41:35 : 14/14  0.0%
18:41:41 : 15/15  0.0%
18:41:45 : 16/16  0.0%
18:41:46 : Test finished.

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

Again, very easy to ABX. -q0 introduces severe distortions to the sound, very 'chewing-gum' and 'bladisch'. And again, the distortions could be admired in a frequency visualization: CBR 128 is very flat, whereas q0 introduce a lot of variations and it breaks the linearity of the lowpass value.

-qo in 3.96

Reply #4 – 2004-08-02 20:01:48

Does it start to appear on -q2?

-qo in 3.96

Reply #5 – 2004-08-02 20:16:34

Tested with Dom Bedos (organ) at CBR 128:
- q0 and q1 produce a nasty ringing
- q2 and q3 are much better.

Bit to bit comparison between q2 and q3 on this sample

Code: [Select]

INFO (foo_bitcompare) : location: "file://C:\temp\HA PACK\__samples\Dom Bedos#3-Q2.mp3" (0)
INFO (foo_bitcompare) : location: "file://C:\temp\HA PACK\__samples\Dom Bedos#3-Q3.mp3" (0)
INFO (foo_bitcompare) : No differences in decoded data found.
INFO (foo_bitcompare) : Finished successfully.

-qo in 3.96

Reply #6 – 2004-08-02 21:37:49

looks like "noise_shaping_amp==2"--which should amplify only one distorted scalefactor band at once--does not work well in cbr conditions. seems we have to disable that.

did some code archaeology. Naoki had this amplification mode in mind as he worked on nspsytune. But I can't remember, if he wanted to use it for vbr only.

-qo in 3.96

Reply #7 – 2004-08-02 22:14:39

Or increase max_age?
Anyway this needs investigation.

Notice