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: I found a really bad killer sample. (Read 61049 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

I found a really bad killer sample.

I've seemed to have come across a track from my flac collection that has a really bad precho problem, just like eig. At the very start of the track it has artifacts and is even ABXable at 320kbps.

LAME 3.98.2

V2

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/21 22:00:17

File A: C:\Rips\Music\Ministry - Side Trax\13. Show Me Your Spine (PTP).flac
File B: C:\Temp\PTP - Show Me Your Spine LAME 3.98 V2.mp3

22:00:17 : Test started.
22:00:47 : 01/01  50.0%
22:00:50 : 02/02  25.0%
22:00:54 : 03/03  12.5%
22:00:58 : 04/04  6.3%
22:01:03 : 05/05  3.1%
22:01:07 : 06/06  1.6%
22:01:12 : 07/07  0.8%
22:01:15 : 08/08  0.4%
22:01:19 : 09/09  0.2%
22:01:23 : 10/10  0.1%
22:01:28 : 11/11  0.0%
22:01:31 : 12/12  0.0%
22:01:37 : Test finished.

 ---------- 
Total: 12/12 (0.0%)

Precho and scratching noises.

V0

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/21 22:03:28

File A: C:\Rips\Music\Ministry - Side Trax\13. Show Me Your Spine (PTP).flac
File B: C:\Temp\PTP - Show Me Your Spine LAME 3.98 V0.mp3

22:03:28 : Test started.
22:03:44 : 01/01  50.0%
22:03:53 : 02/02  25.0%
22:03:58 : 03/03  12.5%
22:04:04 : 04/04  6.3%
22:04:09 : 05/05  3.1%
22:04:16 : 06/06  1.6%
22:04:20 : 07/07  0.8%
22:04:26 : 08/08  0.4%
22:04:30 : 09/09  0.2%
22:04:33 : 10/10  0.1%
22:04:38 : 11/11  0.0%
22:04:42 : 12/12  0.0%
22:04:50 : 13/13  0.0%
22:04:57 : 14/14  0.0%
22:05:02 : 15/15  0.0%
22:05:03 : Test finished.

 ---------- 
Total: 15/15 (0.0%)

A little improvement.

-b 320

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/21 22:08:28

File A: C:\Rips\Music\Ministry - Side Trax\13. Show Me Your Spine (PTP).flac
File B: C:\Temp\PTP - Show Me Your Spine LAME 3.98 320 CBR.mp3

22:08:28 : Test started.
22:08:39 : 01/01  50.0%
22:08:46 : 02/02  25.0%
22:08:50 : 03/03  12.5%
22:08:55 : 04/04  6.3%
22:09:02 : 05/05  3.1%
22:09:07 : 06/06  1.6%
22:09:13 : 07/07  0.8%
22:09:23 : 08/08  0.4%
22:09:30 : 09/09  0.2%
22:09:35 : 10/10  0.1%
22:09:40 : 11/11  0.0%
22:09:44 : 12/12  0.0%
22:09:49 : 13/13  0.0%
22:09:57 : 14/14  0.0%
22:10:07 : 15/15  0.0%
22:10:08 : Test finished.

 ---------- 
Total: 15/15 (0.0%)

A little bit better then V0.

I also tried out some other Mp3 encoders.

iTunes

192kbps VBR High

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/21 22:31:54

File A: C:\Temp\Show Me Your Spine [Sample] iTunes Mp3 192 VBR High.mp3
File B: C:\Temp\Show Me Your Spine [Sample].wav

22:31:54 : Test started.
22:32:03 : 01/01  50.0%
22:32:07 : 02/02  25.0%
22:32:10 : 03/03  12.5%
22:32:14 : 04/04  6.3%
22:32:18 : 05/05  3.1%
22:32:23 : 06/06  1.6%
22:32:27 : 07/07  0.8%
22:32:30 : 08/08  0.4%
22:32:35 : 09/09  0.2%
22:32:38 : 10/10  0.1%
22:32:42 : 11/11  0.0%
22:32:51 : 12/12  0.0%
22:32:52 : Test finished.

 ---------- 
Total: 12/12 (0.0%)

Sounds worse then LAME V2.

320kbps CBR

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/21 22:33:18

File A: C:\Temp\Show Me Your Spine [Sample] iTunes 320 CBR.mp3
File B: C:\Temp\Show Me Your Spine [Sample].wav

22:33:18 : Test started.
22:33:57 : 01/01  50.0%
22:34:01 : 02/02  25.0%
22:34:07 : 03/03  12.5%
22:34:12 : 04/04  6.3%
22:34:16 : 05/05  3.1%
22:34:20 : 06/06  1.6%
22:34:26 : 07/07  0.8%
22:34:30 : 08/08  0.4%
22:34:34 : 09/09  0.2%
22:34:38 : 10/10  0.1%
22:34:42 : 11/11  0.0%
22:34:47 : 12/12  0.0%
22:34:52 : 13/13  0.0%
22:34:57 : 14/14  0.0%
22:35:03 : 15/15  0.0%
22:35:05 : Test finished.

 ---------- 
Total: 15/15 (0.0%)

Sounds better, but alot worse then LAME 320.

FhG Mp3 Surround 1.5

M1 q1

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/21 23:13:04

File A: C:\Temp\Show Me Your Spine Fhg VBR M1.mp3
File B: C:\Temp\Show Me Your Spine [Sample].flac

23:13:04 : Test started.
23:13:16 : 01/01  50.0%
23:13:19 : 02/02  25.0%
23:13:23 : 03/03  12.5%
23:13:26 : 04/04  6.3%
23:13:29 : 05/05  3.1%
23:13:33 : 06/06  1.6%
23:13:38 : 07/07  0.8%
23:13:41 : 08/08  0.4%
23:13:45 : 09/09  0.2%
23:13:52 : 10/10  0.1%
23:13:57 : 11/11  0.0%
23:14:01 : 12/12  0.0%
23:14:02 : Test finished.

 ---------- 
Total: 12/12 (0.0%)

320kbps

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/21 23:14:25

File A: C:\Temp\Show Me Your Spine Fhg 320.mp3
File B: C:\Temp\Show Me Your Spine [Sample].flac

23:14:25 : Test started.
23:14:35 : 01/01  50.0%
23:14:39 : 02/02  25.0%
23:14:42 : 03/03  12.5%
23:14:46 : 04/04  6.3%
23:14:49 : 05/05  3.1%
23:14:51 : 06/06  1.6%
23:14:54 : 07/07  0.8%
23:14:58 : 08/08  0.4%
23:15:03 : 09/09  0.2%
23:15:07 : 10/10  0.1%
23:15:11 : 11/11  0.0%
23:15:16 : 12/12  0.0%
23:15:18 : Test finished.

 ---------- 
Total: 12/12 (0.0%)
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #1 –
I confirm at Lame V2, I wasn't able to take V2 faulty in my own comparison so thx for the sample, I may add it.

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/22 07:40:39

File A: C:\Documents and Settings\aShow_Me_Your_Spine__Sample_.mp3
File B: C:\Documents and Settings\Show_Me_Your_Spine__Sample_.flac

07:40:39 : Test started.
07:40:57 : 01/01  50.0%
07:41:09 : 02/02  25.0%
07:41:41 : 03/03  12.5%
07:41:59 : 04/04  6.3%
07:42:26 : 04/05  18.8%
07:42:51 : 05/06  10.9%
07:43:11 : 06/07  6.3%
07:43:50 : 07/08  3.5%
07:44:32 : 08/09  2.0%
07:44:59 : 09/10  1.1%
07:45:41 : 10/11  0.6%
07:46:10 : 11/12  0.3%
07:46:16 : Test finished.

 ---------- 
Total: 11/12 (0.3%)


one second of this sample is enough I hear a distortion (like sand on a drum) on the third beat exactly, I didn't bother listening to the rest of the sample.
If you have other ghosts like that in the closet i am interested, I will have a look at this eig sample if I find it.

I found a really bad killer sample.

Reply #2 –
heres the eig sample, its pretty bad on just about anything
My $.02, may not be in the right currency

I found a really bad killer sample.

Reply #3 –
waouh this one is terribly easy to ABX, I hear distortions in the little explosions in the background. Thks thundat00th.

Lame V2

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/22 09:06:21

File A: C:\Documents and Settings\aeig.mp3
File B: C:\Documents and Settings\eig.flac

09:06:21 : Test started.
09:06:46 : 01/01  50.0%
09:07:15 : 02/02  25.0%
09:07:49 : 03/03  12.5%
09:08:27 : 04/04  6.3%
09:08:49 : 05/05  3.1%
09:09:20 : 06/06  1.6%
09:09:48 : 07/07  0.8%
09:10:35 : 08/08  0.4%
09:10:37 : Test finished.

 ---------- 
Total: 8/8 (0.4%)


Does anyone know where it comes from Artist/Album ?

PS: I am searching for these ones too Electronic: Spahm (trance music), Amnesia (acid music), I heard pio2001 speaking about these.

I found a really bad killer sample.

Reply #5 –
Thks lvqcl,
as soon as I get some time I will add these two samples to my comparison, because both illustrate that lame is not transparent as V2 which my table doesn't actually show. I also need to see if Nero q0.55 still achieve transparency on these two really bad ones.

I found a really bad killer sample.

Reply #6 –
Quote from: sauvage78 on
If you have other ghosts like that in the closet i am interested, I will have a look at this eig sample if I find it.


Theres a album track by Skinny Puppy that has some bad precho problems at the start with most encoders at low and mid bitrates. Also Linchpin by Fear Factory brings Nero AAC to its knees.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #7 –
Huston we have a problem.

I tried to edit this sample in order to make it shorter & focus on the 4 sec (specially the third beat) that I can ABX & I had a big surprise: when encoded to lame V7, the artefact moved within the sample 

If you encode the original sample to V7, you have a big artefact at the start, on the third beat.

If you encode my shortened version of the sample, you have a big Artefact at the end of the sample that is not here in the long version of the same sample encoded at the same lame setting (V7).

First I thought that I cutted the sample at the wrong point with audacity, but I wasn't the case. Then I thought that the artefact was masked by the following sound, but it wasn't the case too.
I was so surprized that I decoded to wav the long sample encoded to V7 from the original & I edited it to match my short V7 version.

I discovered that the original & the shortened sample didn't react to lame V7 the same way at all despite the fact that the original audio data is the same.

Within the original 30 sec sample the major artefact is at the beginning.

Within the edited 4 sec sample the original artefact is here but much softer & there is an additionnal major artefact at sec 3-4 that is not here if you encode the original to V7.

That is very very strange. I dunno what to do with this sample for my own listening test.

If each time you edit a sample the encoder reacts a different way, this is a problem for any ABXing tests.

I dunno if this is a lame bug or if this is normal. Some people told me to use longer sample, but that doesn't change anything, it seems that with lame each edited version will "maybe" react differently.
So you have to compare the original version Vs. the encoded version AND the edited original version Vs. the edited encoded version ... I doubt anyone did it in previous listening test.
I hope this is a lame bug because otherwise this is very annoying for any listening test. I didn't test on other codecs to see how they react.

Well I can still ABX the same kind of artefact on this sample, so in the end it does make a killer sample a normal sample, but the fact that this soften the first artefact is a real problem.

  I am completely perplex.

Edit1: There was some silence in the beginning of the original I will edit the long version without removing the silence in the beginning (only cutting the end) & see how it reacts ...
Edit2: It reacts as the original if I don't cut the 0.35 msec of silence from the beginning ... so it seems you can cut the end with no problem but not the beginning ... I was warn but I didn't thought it was so severe ... because it affects music at up to 4 sec after the beginning ...
Edit3: Silence or not it shouldn't be different. It is still strange, because it's music that is supposed to affect the model, not silence.
Edit4: I tried to add 35 millisecond of artificial silence to the short edited version. I end with a new beast: even softer artefact in the beginning & no major artefact at the end, it's making me crazy
Edit5: I copy/pasted the sample to have it twice (X2) in a row within the same sample, the result is (soften artefact+major artefact)+(soften artefact+no artefact). Seems the longer is the sample the softer is the artefact. But you cannot say the more there is silence in the begining the softer the artefact. Very hard to predict.

Edit6: Deleted attached files, attachement limit reached.

I found a really bad killer sample.

Reply #8 –
@ sauvage78

Your right about that major artifact moving to the end.

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/24 13:58:17

File A: C:\Downloads\The_Artefact_That_Moves\The Artefact That Moves\03- Show Me Your Spine - Lossless Short.flac
File B: C:\Downloads\The_Artefact_That_Moves\The Artefact That Moves\04- Show Me Your Spine - V7 Short (Major Artefact at the END+Softer Artefact at the START).mp3

13:58:17 : Test started.
13:58:32 : 01/01  50.0%
13:58:35 : 02/02  25.0%
13:58:38 : 03/03  12.5%
13:58:41 : 04/04  6.3%
13:58:44 : 05/05  3.1%
13:58:49 : 06/06  1.6%
13:58:54 : 07/07  0.8%
13:59:02 : 08/08  0.4%
13:59:06 : 09/09  0.2%
13:59:13 : 10/10  0.1%
13:59:22 : 11/11  0.0%
13:59:26 : 12/12  0.0%
13:59:28 : Test finished.

 ---------- 
Total: 12/12 (0.0%)

It appears at around 0:03 - 0:04, instead of the first second.

I think this might be a block usage detection issue, which the mp3 codec can have limitations on with certain samples.

This problem is almost the same problem i have with this track thats not transparent on LAME V0 but sounds fine on V2 

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/24 14:03:14

File A: C:\Temp\LAME V0\Ministry - Houses Of The Molé\01. No W.mp3
File B: C:\Rips\Music\Ministry - Houses Of The Molé\01. No W.flac

14:03:14 : Test started.
14:03:32 : 01/01  50.0%
14:03:37 : 02/02  25.0%
14:03:43 : 03/03  12.5%
14:03:50 : 04/04  6.3%
14:03:58 : 05/05  3.1%
14:04:02 : 06/06  1.6%
14:04:09 : 07/07  0.8%
14:04:14 : 08/08  0.4%
14:04:19 : 09/09  0.2%
14:04:24 : 10/10  0.1%
14:04:25 : Test finished.

 ---------- 
Total: 10/10 (0.1%)

Smear and warbling at 2:42. Shares the same problems with Die in a Crash, which is also easy to ABX at V0 while it sounds better with V2.

Then i decide to make a sample of this track:

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/24 14:06:31

File A: C:\Temp\Ministry - No W [Edit].wav
File B: C:\Temp\Ministry - No W [Edit] V0.mp3

14:06:31 : Test started.
14:07:09 : 00/01  100.0%
14:07:14 : 01/02  75.0%
14:07:19 : 02/03  50.0%
14:07:25 : 02/04  68.8%
14:07:35 : 03/05  50.0%
14:07:40 : 03/06  65.6%
14:07:45 : 04/07  50.0%
14:07:50 : 04/08  63.7%
14:07:54 : 04/09  74.6%
14:07:59 : 04/10  82.8%
14:08:05 : 05/11  72.6%
14:08:15 : 05/12  80.6%
14:08:16 : Test finished.

 ---------- 
Total: 5/12 (80.6%)

Artifact is not present at all.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #9 –
@sauvage78:
You might consider removing multiples of 576 samples (if no resampling is involved) or multiples of 1152 samples (if LR/MS stereo mode switching is critical).

I found a really bad killer sample.

Reply #10 –
I made another edit of the hard to encode track at 3:00  - 3:20 were it sounds really bad. Am shocked that a single track can have soo many artifacts that is even audiable at 320kbps.

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.4
2009/04/04 00:49:37

File A: C:\Temp\PTP_Sample2.flac
File B: C:\Temp\PTP Sample 2 b 320.mp3

00:49:37 : Test started.
00:50:04 : 01/01  50.0%
00:50:14 : 02/02  25.0%
00:50:23 : 03/03  12.5%
00:50:28 : 04/04  6.3%
00:50:34 : 05/05  3.1%
00:50:42 : 06/06  1.6%
00:50:48 : 07/07  0.8%
00:50:54 : 08/08  0.4%
00:51:01 : 09/09  0.2%
00:51:08 : 10/10  0.1%
00:51:16 : 11/11  0.0%
00:51:23 : 12/12  0.0%
00:51:26 : Test finished.

 ---------- 
Total: 12/12 (0.0%)

Encoded with LAME 3.98.2 at 320 kbps CBR. Major precho artifacts.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #11 –
Temp upload for this post.

Edit

I added a few more temp uploads to report a bug on foo_gep.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #12 –
I tried out Vista again and spent about a few hours just to install Service Pack 2 and a shed load of restarts (I've should have made a streamlined copy instead). After playing music foobar2000 under Vista and notice this sample sounds bad on Vista's mixer but it sounds fine if I use foobar's WASAPI output plugin. But am not sure how to ABX the Software mixer and the WASAPI output together, to proof my claim.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #13 –
Quote from: sauvage78 on
PS: I am searching for these ones too Electronic: Spahm (trance music), Amnesia (acid music), I heard pio2001 speaking about these.


Hello,

Here they are :
Amnesia
Spahm

Beware that Amnesia has been normalized. Check for internal clipping when you make an mp3 of it.

I recently ABXed spahm in mp3 at 320 kbps, with Lame 3.98.2.

I didn't try Amnesia. Originally, it was a problem with Musepack more than mp3. The sample is semi-artificial. The original has a much lower volume and causes no problem with Musepack.

I found a really bad killer sample.

Reply #14 –
A regression sample that effects Nero AAC at 0.55 from 0.50.

"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #15 –
Another sample for Nero AAC
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #16 –
I've uploaded some samples from Kraftwerk's epic album, The Man Machine. Which is very easy to ABX at 320kbps with Mp3.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #17 –
An upload for this topic.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #19 –
Uploads for this topic.

Those samples are ABXable at 320kbps Mp3 to those that are sensitive to pre-echo.

Please note, that ptp_robocop.flac is a edit version of Show Me Your Spine (3:00 - 3:30) and the_robots.flac is a different edit from orginal the kraftwerk.wv sample from a recent remsatered release.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #20 –
Another sample that has obvious artifacts at 320 CBR:

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.1
2010/09/13 04:33:48

File A: C:\Temp\Boing Boom Tschak.flac
File B: C:\Temp\Boing Boom Tschak 320 CBR.mp3

04:33:48 : Test started.
04:34:11 : 01/01  50.0%
04:34:13 : 02/02  25.0%
04:34:17 : 03/03  12.5%
04:34:21 : 04/04  6.3%
04:34:25 : 05/05  3.1%
04:34:28 : 06/06  1.6%
04:34:32 : 07/07  0.8%
04:34:35 : 08/08  0.4%
04:34:38 : 09/09  0.2%
04:34:41 : 10/10  0.1%
04:34:44 : 11/11  0.0%
04:34:48 : 12/12  0.0%
04:34:51 : 13/13  0.0%
04:34:53 : 14/14  0.0%
04:34:56 : 15/15  0.0%
04:35:00 : 16/16  0.0%
04:35:04 : 17/17  0.0%
04:35:07 : 18/18  0.0%
04:35:09 : 19/19  0.0%
04:35:12 : 20/20  0.0%
04:35:14 : 21/21  0.0%
04:35:17 : 22/22  0.0%
04:35:20 : 23/23  0.0%
04:35:24 : 24/24  0.0%
04:35:29 : Test finished.

 ---------- 
Total: 24/24 (0.0%)

The sample is for this subject.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #21 –
An re-upload for this topic.

Most of those samples on this zip file are also ABXable on modern codecs as well.  For example Ogg Vorbis fails badly at the start on The Robots.
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #22 –
A couple more samples that is ABXable at 320kbps:

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.1.7
2011/07/05 20:14:48

File A: C:\Temp\Home Computer.flac
File B: C:\Temp\Home Computer.mp3

20:14:48 : Test started.
20:15:01 : 01/01  50.0%
20:15:08 : 02/02  25.0%
20:15:11 : 03/03  12.5%
20:15:16 : 04/04  6.3%
20:15:20 : 05/05  3.1%
20:15:25 : 06/06  1.6%
20:15:32 : 07/07  0.8%
20:15:40 : 08/08  0.4%
20:15:48 : 09/09  0.2%
20:15:52 : 10/10  0.1%
20:15:57 : 11/11  0.0%
20:16:08 : 12/12  0.0%
20:16:10 : Test finished.

 ---------- 
Total: 12/12 (0.0%)

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v1.1.7
2011/07/05 20:17:35

File A: C:\Temp\That's What I Get.flac
File B: C:\Temp\That's What I Get.mp3

20:17:35 : Test started.
20:17:43 : 01/01  50.0%
20:17:49 : 02/02  25.0%
20:17:58 : 03/03  12.5%
20:18:13 : 04/04  6.3%
20:18:20 : 05/05  3.1%
20:18:29 : 06/06  1.6%
20:18:37 : 07/07  0.8%
20:18:45 : 08/08  0.4%
20:18:58 : 09/09  0.2%
20:19:07 : 10/10  0.1%
20:19:16 : 11/11  0.0%
20:19:28 : 12/12  0.0%
20:19:29 : Test finished.

 ---------- 
Total: 12/12 (0.0%)
"I never thought I'd see this much candy in one mission!"

I found a really bad killer sample.

Reply #23 –
I am so glad I can't hear these differences.
Does it mean that I can't hear anything else and rest in peace in the lossy realm?

I found a really bad killer sample.

Reply #24 –
/mnt is very sensitive to pre-echo problems, much more than most of us.
This is very good for getting experience with codecs' issues when using very high bitrate, but you shouldn't care if you can't hear issues /mnt can hear.
lame3995o -Q1.7 --lowpass 17