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: A strange artifact on LAME (Read 7179 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

A strange artifact on LAME

I just did a quick ABX on a track that I thought sounded it had a artifact, then after getting 16/22 at V2, I then tried out V0 to make it harder and for some reason V0 was alot more easy to ABX.

LAME 3.97 -V0 --vbr-new

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.5.4
2008/08/02 01:13:43

File A: C:\Temp\Dope Hat.mp3
File B: C:\Temp\Listen Tests\06. Dope Hat.flac

01:13:43 : Test started.
01:14:43 : 01/01  50.0%
01:14:53 : 01/02  75.0%
01:15:04 : 02/03  50.0%
01:15:19 : 03/04  31.3%
01:15:23 : 04/05  18.8%
01:15:39 : 05/06  10.9%
01:15:46 : 06/07  6.3%
01:16:28 : 07/08  3.5%
01:16:37 : 08/09  2.0%
01:16:56 : 09/10  1.1%
01:17:17 : 10/11  0.6%
01:17:29 : 11/12  0.3%
01:17:44 : 12/13  0.2%
01:18:00 : 13/14  0.1%
01:18:11 : 14/15  0.0%
01:18:23 : Test finished.

 ---------- 
Total: 14/15 (0.0%)

For some reason some of sound on the drum snare at 2:54 is gone, giving it a cleaner sound. Which gave me the impression that file B was lossy.

LAME 3.97 -V 2 --vbr-new

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.5.4
2008/08/02 01:24:15

File A: C:\Music\Albums\Marilyn Manson - Portrait Of An American Family\06. Dope Hat.mp3
File B: C:\Temp\Listen Tests\06. Dope Hat.flac

01:24:15 : Test started.
01:24:52 : 01/01  50.0%
01:25:02 : 02/02  25.0%
01:25:35 : 03/03  12.5%
01:26:06 : 04/04  6.3%
01:26:31 : 05/05  3.1%
01:26:36 : 05/06  10.9%
01:26:49 : 05/07  22.7%
01:26:56 : 05/08  36.3%
01:27:53 : 06/09  25.4%
01:27:56 : 06/10  37.7%
01:28:08 : 06/11  50.0%
01:28:38 : 07/12  38.7%
01:28:42 : Test finished.

 ---------- 
Total: 7/12 (38.7%)

A had trouble at V2 at around 2:54 unlike V0, but I have mix luck and hearing at that setting but it was alot harder to ABX at V2.

Also I did find a artifact like this a few months ago with LAME 3.90.3 at preset standard on a NIN track called "Ruiner", which a certain pitch noise in the vocals on the track was missing, which gave a cleaner feel.

Code: [Select]
foo_abx 1.3.1 report
foobar2000 v0.9.5.1 beta 1
2008/02/22 22:04:50

File A: E:\Music\Albums\Nine Inch Nails - The Downward Spiral\06. Ruiner.mp3
File B: E:\Temp\Listen Tests\Ruiner.flac

22:04:50 : Test started.
22:05:10 : 01/01  50.0%
22:05:41 : 02/02  25.0%
22:06:07 : 03/03  12.5%
22:06:54 : 04/04  6.3%
22:07:49 : 05/05  3.1%
22:09:30 : 06/06  1.6%
22:09:38 : 07/07  0.8%
22:10:23 : 08/08  0.4%
22:10:40 : 09/09  0.2%
22:12:04 : 10/10  0.1%
22:12:22 : 11/11  0.0%
22:12:23 : Test finished.

 ---------- 
Total: 11/11 (0.0%)
"I never thought I'd see this much candy in one mission!"

A strange artifact on LAME

Reply #1 –
It disturbs me that you don't ABX a fixed number of rounds.
Perhaps "disturbs" is the wrong word, but why do you not pick a number of rounds and stick with it for all testing?
Are you watching the results and adjusting the number of rounds "on the fly"?
Creature of habit.

A strange artifact on LAME

Reply #2 –
Quote from: Soap on
It disturbs me that you don't ABX a fixed number of rounds.
Perhaps "disturbs" is the wrong word, but why do you not pick a number of rounds and stick with it for all testing?
Are you watching the results and adjusting the number of rounds "on the fly"?
I did 2 rounds on those ABX tests, those results are from my second test.

Anyway I did another round but on ABC Hidden Reference this time, which makes the test more harder.

LAME 3.97 -V0 --vbr-new

Code: [Select]
1 of   1, p = 0.500
  2 of   2, p = 0.250
  2 of   3, p = 0.500
  3 of   4, p = 0.313
  4 of   5, p = 0.188
  5 of   6, p = 0.109
  5 of   7, p = 0.227
  6 of   8, p = 0.145
  7 of   9, p = 0.090
  7 of  10, p = 0.172
  7 of  11, p = 0.274
  8 of  12, p = 0.194
  9 of  13, p = 0.133
 10 of  14, p = 0.090
 11 of  15, p = 0.059
 12 of  16, p = 0.038
 13 of  17, p = 0.025
 14 of  18, p = 0.015
 15 of  19, p = 0.010
 16 of  20, p = 0.006

ABC/HR Version 1.0, 6 May 2004
Testname: 

1L = C:\Temp\Dope Hat V0.wav

---------------------------------------
General Comments:

---------------------------------------
ABX Results:
Original vs C:\Temp\Dope Hat V0.wav
    16 out of 20, pval = 0.006
"I never thought I'd see this much candy in one mission!"

A strange artifact on LAME

Reply #3 –
I did another round again after having a good sleep, to see if the results would be any better, but got 10/14 with ABC and 11/12 with foobar2000.

Any I have edited the V0 encoded file and the lossless source file and have uploaded, if anyone would like to see if they can ABX it. The download link is just below and above the ABX log.

Dope Hat Sample

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.5.4
2008/08/02 12:44:52

File A: C:\Temp\Dope Hat [Edit] LAME3.97 V0.wav
File B: C:\Temp\Dope Hat [Edit].wav

12:44:52 : Test started.
12:45:25 : 00/01  100.0%
12:45:45 : 01/02  75.0%
12:45:57 : 02/03  50.0%
12:46:21 : 03/04  31.3%
12:46:24 : 04/05  18.8%
12:46:36 : 05/06  10.9%
12:46:39 : 06/07  6.3%
12:46:52 : 07/08  3.5%
12:46:58 : 08/09  2.0%
12:47:39 : 09/10  1.1%
12:47:44 : 10/11  0.6%
12:48:01 : 11/12  0.3%
12:48:03 : Test finished.

 ---------- 
Total: 11/12 (0.3%)

The artifact is at around 0:24 on this edit.
"I never thought I'd see this much candy in one mission!"

A strange artifact on LAME

Reply #4 –
I managed to do a test with a LAME 3.98 -V0 encode with this track.

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.5.4
2008/08/02 14:17:38

File A: C:\Temp\DopeHat_LAME3.98V0.mp3
File B: C:\Temp\Listen Tests\06. Dope Hat.flac

14:17:38 : Test started.
14:18:47 : 01/01  50.0%
14:19:23 : 01/02  75.0%
14:19:29 : 02/03  50.0%
14:19:36 : 03/04  31.3%
14:20:30 : 03/05  50.0%
14:20:34 : 03/06  65.6%
14:21:01 : 04/07  50.0%
14:21:15 : 04/08  63.7%
14:21:42 : 05/09  50.0%
14:21:50 : 05/10  62.3%
14:21:59 : 06/11  50.0%
14:22:07 : 06/12  61.3%
14:22:41 : 06/13  70.9%
14:22:56 : 07/14  60.5%
14:23:01 : Test finished.

 ---------- 
Total: 7/14 (60.5%)

I got confused with both files on this test at 2:54, but I will to try to ABX it again at some other time.
"I never thought I'd see this much candy in one mission!"

A strange artifact on LAME

Reply #5 –
I also found the odd sample where a higher quality setting introduce a new artifact. Its really weird so I found an old thread asking 'can --preset extreme be worse than standard ?'

Dibrom explained: Its possible due to quantisation error but such events are rare. A higher -V setting has less overall distortion but there is a very slight chance of a new artifact occuring on the higher setting.

A strange artifact on LAME

Reply #6 –
I once had a strange experience with the birds sample also. Must be back some years now but i remember using V2 back then introduced the typical distortion noise. Using higher bitrate made it a bit less but added a strange click alike plop sound
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

A strange artifact on LAME

Reply #7 –
Quote from: shadowking on
I also found the odd sample where a higher quality setting introduce a new artifact. Its really weird so I found an old thread asking 'can --preset extreme be worse than standard ?'

Dibrom explained: Its possible due to quantisation error but such events are rare. A higher -V setting has less overall distortion but there is a very slight chance of a new artifact occuring on the higher setting.
I see, but I do have a few tracks in my music collection, that sometimes a higher -V setting produces a artifact. Such as Back The F**k Up by Fear Factory, which was a smear artifact at -V1 while V2 it sounds better, and Run To The Hills by Iron Maiden has a smear at 0:48 on V2, while I found it transparent at V3.

LAME 3.97 -V3 (A) --vbr-new vs -V2 (B) --vbr-new

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.5.5
2008/08/05 14:03:37

File A: C:\Temp\LAME V3\Iron Maiden - The Number Of The Beast\06. Run To The Hills.mp3
File B: C:\Music\Albums\Iron Maiden - The Number Of The Beast\06. Run To The Hills.mp3

14:03:37 : Test started.
14:04:00 : 01/01  50.0%
14:04:05 : 02/02  25.0%
14:04:32 : 03/03  12.5%
14:05:11 : 04/04  6.3%
14:05:36 : 05/05  3.1%
14:05:52 : 06/06  1.6%
14:06:18 : 07/07  0.8%
14:06:29 : 08/08  0.4%
14:06:34 : 09/09  0.2%
14:06:43 : 10/10  0.1%
14:07:03 : Test finished.

 ---------- 
Total: 10/10 (0.1%)

A smear or a precho at 0:48 on file B (-V2 --vbr-new), anyway the track seems fine on 3.98 at V2 though.

Quote from: Wombat on
I once had a strange experience with the birds sample also. Must be back some years now but i remember using V2 back then introduced the typical distortion noise. Using higher bitrate made it a bit less but added a strange click alike plop sound
Is the birds sample has that sandpaper like distortion sound like on the trumpet sample does?
"I never thought I'd see this much candy in one mission!"

A strange artifact on LAME

Reply #8 –
Its similar to my findings. Could the older pre 3.98 vbr-new cause it ?. Its been tuned heavily recently. I once posted two samples where v3 introduced something new but not v2 and not v4 -BUT vbr-old was fine. This was with 3.97 and early 3.98b. The last 3.98 builds were good.

A strange artifact on LAME

Reply #9 –
Quote from: shadowking on
Its similar to my findings. Could the older pre 3.98 vbr-new cause it ?. Its been tuned heavily recently. I once posted two samples where v3 introduced something new but not v2 and not v4 -BUT vbr-old was fine. This was with 3.97 and early 3.98b. The last 3.98 builds were good.

I have got a track that --vbr-new was not transparent on, and it was fine on --vbr-old and 192 CBR. I think its transparent on 3.98 now, but I will have to a ABX on it at some other time.
"I never thought I'd see this much candy in one mission!"