A sample for a track that I used on a test (http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=67840&view=findpost&p=603810) that has a noticable precho artifact that produces a raining noise in the background.
I can't hear the rain noise on my crappy laptop speakers but I sure do love Skinny Puppy
If you can post a FLAC of the wav clip too that would be great that way I can ABX it.
If you can post a FLAC of the wav clip too that would be great that way I can ABX it.
That FLAC is of the original.
FWIW I can't ABX lame 3.97 V2 vbr-new
Cheers,
David.
Maybe it my hardware (SB Audigy SE + Sen 215 HD) producing the artifact, since some of the Sound Blaster sound cards have known to have a crappy DSP that sucks at resampling. Anyway i will upgrade my sound card around Christmas or early next year, but atm cannot decide to go with a ASUS Xonar DX1 (PCI) or a X-Fi Extreme Gamer (hardware mixing on Linux would be really nice).
Am sure someone else can ABX this sample aswell. Anyway i did a ABX with the sample at V2 --vbr-new with LAME 3.97.
foo_abx 1.3.3 report
foobar2000 v0.9.6
2008/12/10 11:26:25
File A: C:\Temp\Human_Disease V2.mp3
File B: C:\Temp\Human_Disease.flac
11:26:25 : Test started.
11:26:56 : 01/01 50.0%
11:27:06 : 02/02 25.0%
11:27:16 : 03/03 12.5%
11:27:25 : 04/04 6.3%
11:27:38 : 05/05 3.1%
11:27:52 : 06/06 1.6%
11:28:04 : 07/07 0.8%
11:28:13 : 08/08 0.4%
11:28:19 : 09/09 0.2%
11:28:45 : 10/10 0.1%
11:29:06 : 11/11 0.0%
11:29:23 : 12/12 0.0%
11:29:24 : Test finished.
----------
Total: 12/12 (0.0%)
Sounds watery at around 0:01 - 0:03, at around 0:02 its noticable.
Maybe it my hardware (SB Audigy SE + Sen 215 HD) producing the artifact, since some of the Sound Blaster sound cards have known to have a crappy DSP that sucks at resampling.
I use fb2k's PPHS resampler in ultra mode when performing ABX tests. Regarding SB's crappy resampling, my Audigy (the oldest one with a firewire port) seems to be worse than my dead SB Live based on these tests:
http://www.hydrogenaudio.org/forums/index.php?showtopic=9772 (http://www.hydrogenaudio.org/forums/index.php?showtopic=9772)
http://www.hydrogenaudio.org/forums/index....showtopic=65398 (http://www.hydrogenaudio.org/forums/index.php?showtopic=65398)
I brought a Sound Blaster X-Fi Xtreme Gamer soundcard this week, and decided to see if using a higher end soundcard might improve this sample. So far it's a nice upgrade; i got the Intel HD Audio case connectors working on it and the new GPL Linux drivers works fine with Ubuntu 8.10 with hardware mixing , i can now play quake 1-3 and rtcw with a alsa or oss music app playing in the background.
LAME 3.97 -2 --vbr-newfoo_abx 1.3.3 report
foobar2000 v0.9.6.1
2009/02/27 21:03:53
File A: C:\Downloads\Human_Disease.flac
File B: C:\Temp\1 Human_Disease.mp3
21:03:53 : Test started.
21:04:30 : 01/01 50.0%
21:04:37 : 02/02 25.0%
21:04:43 : 03/03 12.5%
21:04:50 : 04/04 6.3%
21:04:59 : 05/05 3.1%
21:05:08 : 06/06 1.6%
21:05:22 : 07/07 0.8%
21:05:36 : 08/08 0.4%
21:05:50 : 09/09 0.2%
21:05:59 : 10/10 0.1%
21:06:08 : 11/11 0.0%
21:06:20 : 12/12 0.0%
21:06:26 : Test finished.
----------
Total: 12/12 (0.0%)
Sadly i can still hear the artifact, even if i haven't performed a ABX test in months. The first 5 secs has a metallic noise that produces a wet effect.
LAME 3.97 -V0 --vbr-newfoo_abx 1.3.3 report
foobar2000 v0.9.6.1
2009/02/27 21:13:50
File A: C:\Downloads\Human_Disease.flac
File B: C:\Temp\1 Human_Disease V0.mp3
21:13:50 : Test started.
21:14:10 : 01/01 50.0%
21:14:37 : 02/02 25.0%
21:14:48 : 03/03 12.5%
21:15:07 : 04/04 6.3%
21:15:21 : 05/05 3.1%
21:15:40 : 06/06 1.6%
21:15:59 : 07/07 0.8%
21:16:06 : 08/08 0.4%
21:16:13 : 09/09 0.2%
21:16:33 : 10/10 0.1%
21:16:59 : 11/11 0.0%
21:17:11 : 12/12 0.0%
21:17:12 : Test finished.
----------
Total: 12/12 (0.0%)
An odd audible noise at over 0:02.
LAME 3.98 -V2foo_abx 1.3.3 report
foobar2000 v0.9.6.1
2009/02/27 21:22:35
File A: C:\Downloads\Human_Disease.flac
File B: C:\Temp\1 Human_Disease v2 3.98.mp3
21:22:35 : Test started.
21:22:53 : 01/01 50.0%
21:22:57 : 02/02 25.0%
21:23:04 : 03/03 12.5%
21:23:19 : 04/04 6.3%
21:23:26 : 05/05 3.1%
21:23:35 : 06/06 1.6%
21:23:43 : 07/07 0.8%
21:23:48 : 08/08 0.4%
21:23:55 : 09/09 0.2%
21:24:25 : 10/10 0.1%
21:24:31 : 11/11 0.0%
21:24:37 : 12/12 0.0%
21:24:38 : Test finished.
----------
Total: 12/12 (0.0%)
The artifact also appears with LAME 3.98 sounds more metallic.
I tested the sample with other codecs aswell. It seems that even ogg, mpc and aac struggle with this sample on standard settings.
Musepack 1.16 --standardfoo_abx 1.3.3 report
foobar2000 v0.9.6.1
2009/02/28 18:56:21
File A: C:\Temp\Human_Disease MPC 1.16 SV7 q7.mpc
File B: C:\Downloads\Human_Disease.flac
18:56:21 : Test started.
18:56:48 : 01/01 50.0%
18:56:58 : 02/02 25.0%
18:57:25 : 03/03 12.5%
18:57:51 : 04/04 6.3%
18:57:58 : 05/05 3.1%
18:58:08 : 06/06 1.6%
18:58:20 : 07/07 0.8%
18:58:33 : 08/08 0.4%
18:58:45 : 09/09 0.2%
18:58:53 : 10/10 0.1%
18:59:00 : 11/11 0.0%
18:59:07 : 12/12 0.0%
18:59:26 : 13/13 0.0%
18:59:32 : 14/14 0.0%
18:59:42 : 15/15 0.0%
18:59:44 : Test finished.
----------
Total: 15/15 (0.0%)
Metallic noises at the start, just like the LAME encodings. Please note that "q7" on the filename is a typo it should've been "q5".
Nero AAC 1.3.3.0 -q 0.5foo_abx 1.3.3 report
foobar2000 v0.9.6.1
2009/02/28 19:07:13
File A: C:\Downloads\Human_Disease.flac
File B: C:\Temp\Human_Disease.mp4
19:07:13 : Test started.
19:07:34 : 01/01 50.0%
19:07:40 : 02/02 25.0%
19:07:52 : 03/03 12.5%
19:08:01 : 04/04 6.3%
19:08:15 : 05/05 3.1%
19:08:27 : 06/06 1.6%
19:08:37 : 07/07 0.8%
19:09:00 : 08/08 0.4%
19:09:06 : 09/09 0.2%
19:09:14 : 10/10 0.1%
19:09:22 : 11/11 0.0%
19:09:33 : 12/12 0.0%
19:09:39 : 13/13 0.0%
19:09:40 : Test finished.
----------
Total: 13/13 (0.0%)
Metallic noises at the start aswell.
Nero AAC 1.3.3.0 -q 0.55foo_abx 1.3.3 report
foobar2000 v0.9.6.1
2009/02/28 19:10:13
File A: C:\Downloads\Human_Disease.flac
File B: C:\Temp\Human_Disease q 55.mp4
19:10:13 : Test started.
19:10:50 : 01/01 50.0%
19:11:03 : 02/02 25.0%
19:11:15 : 02/03 50.0%
19:11:26 : 03/04 31.3%
19:11:38 : 04/05 18.8%
19:11:49 : 05/06 10.9%
19:12:11 : 05/07 22.7%
19:12:22 : 05/08 36.3%
19:12:35 : 06/09 25.4%
19:12:50 : 06/10 37.7%
19:12:57 : 07/11 27.4%
19:13:09 : 08/12 19.4%
19:13:17 : 09/13 13.3%
19:13:27 : 10/14 9.0%
19:13:31 : 11/15 5.9%
19:13:38 : 11/16 10.5%
19:13:43 : 11/17 16.6%
19:13:45 : Test finished.
----------
Total: 11/17 (16.6%)
Sometimes i can the hear the metallic noise, but it was hard to notice.
Vorbis aoTuV b5 -q 5foo_abx 1.3.3 report
foobar2000 v0.9.6.1
2009/02/28 19:36:01
File A: C:\Downloads\Human_Disease.flac
File B: C:\Temp\Human_Disease.ogg
19:36:01 : Test started.
19:36:25 : 01/01 50.0%
19:36:36 : 02/02 25.0%
19:36:50 : 02/03 50.0%
19:36:53 : 03/04 31.3%
19:37:03 : 04/05 18.8%
19:37:17 : 05/06 10.9%
19:37:27 : 06/07 6.3%
19:37:34 : 07/08 3.5%
19:37:40 : 08/09 2.0%
19:37:47 : 09/10 1.1%
19:37:58 : 10/11 0.6%
19:38:03 : 11/12 0.3%
19:38:12 : 11/13 1.1%
19:38:22 : 12/14 0.6%
19:38:29 : 13/15 0.4%
19:38:35 : 14/16 0.2%
19:38:45 : 15/17 0.1%
19:38:52 : 16/18 0.1%
19:38:59 : 17/19 0.0%
19:39:10 : 18/20 0.0%
19:39:16 : Test finished.
----------
Total: 18/20 (0.0%)
Maybe q5 was abit too low for vorbis, anyway the metallic noise is present on this encode aswell, but it does sound different from other codecs. Just wondering if this could be a killer sample.
Very interesting findings. Would it be possible that you try the following sample? It's Quicktime 7.6 true VBR AAC.
[attachment=4921:Human_Disease.m4a]
@/mnt
Creative advertizes the X-Fi cards with some "Crystalizer" feature to "enhance" MP3s. Maybe that feature has some side effects? (Or maybe any of the other post processing features they are proud of?)
Very interesting findings. Would it be possible that you try the following sample? It's Quicktime 7.6 true VBR AAC.
[attachment=4921:Human_Disease.m4a]
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/02/28 23:07:12
File A: C:\Downloads\Human_Disease.flac
File B: C:\Downloads\Human_Disease.m4a
23:07:12 : Test started.
23:07:43 : 01/01 50.0%
23:07:53 : 02/02 25.0%
23:07:59 : 03/03 12.5%
23:08:09 : 04/04 6.3%
23:08:23 : 05/05 3.1%
23:08:28 : 06/06 1.6%
23:08:41 : 07/07 0.8%
23:09:16 : 08/08 0.4%
23:09:26 : 09/09 0.2%
23:09:30 : 10/10 0.1%
23:09:43 : 11/11 0.0%
23:09:57 : 12/12 0.0%
23:09:58 : Test finished.
----------
Total: 12/12 (0.0%)
The start of the sample sounds clean, no metallic noises at all. But at over 0.13 the disortion sounds different.
UPDATE: I have performed the test again since i found a noticable artifact on the Quicktime encoded file.
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/02/28 23:28:11
File A: C:\Downloads\Human_Disease.flac
File B: C:\Downloads\Human_Disease.m4a
23:28:11 : Test started.
23:28:45 : 01/01 50.0%
23:28:51 : 02/02 25.0%
23:28:58 : 03/03 12.5%
23:29:16 : 04/04 6.3%
23:29:25 : 05/05 3.1%
23:29:47 : 06/06 1.6%
23:29:56 : 07/07 0.8%
23:30:06 : 08/08 0.4%
23:30:18 : 09/09 0.2%
23:30:30 : 10/10 0.1%
23:30:37 : 11/11 0.0%
23:30:49 : 12/12 0.0%
23:30:50 : Test finished.
----------
Total: 12/12 (0.0%)
At around 0:25 - 0:28 sounds weak and theres a artifact at 0:27.
@/mnt
Creative advertizes the X-Fi cards with some "Crystalizer" feature to "enhance" MP3s. Maybe that feature has some side effects?
I have the Crystalizer feature disabled, since it causes distortion with certain music.
EDIT: Added ABX results
I have tried to repeat your tests. LAME 3.98.2 is instantly ABXable for me up to 145 kbit/s (V5). I have yet to go higher.
At first I couldn't find anything you have talked about in the signal. For everybody wanting to repeat these tests I recommend starting really low (as V7) to find what you have to look out for. I don't agree that the artifact is a raining noise. The original FLAC contains a raining noise starting at about 0.7 (like heavy rain hitting a street) and this noise is missing in the LAME encode (there's only some kind of pink noise left).
I'll try to work up to V2 now.
foo_abx 1.3.3 report
foobar2000 v0.9.6.2
2009/03/05 22:06:32
File A: C:\windows\profiles\crossover\Desktop\My Mac Desktop\Human_Disease.flac
File B: C:\windows\profiles\crossover\Desktop\My Mac Desktop\Human_Disease.mp3
22:06:32 : Test started.
22:07:52 : 01/01 50.0%
22:08:14 : 02/02 25.0%
22:09:24 : 03/03 12.5%
22:09:38 : 04/04 6.3%
22:09:55 : 05/05 3.1%
22:10:07 : 06/06 1.6%
22:10:24 : 07/07 0.8%
22:10:34 : Test finished.
----------
Total: 7/7 (0.8%)
I couldn't find any difference for Nero and Quicktime AAC, yet. Could you elaborate further what you are hearing?
Edit/PS:Well, I finally made it up to V2:
foo_abx 1.3.3 report
foobar2000 v0.9.6.2
2009/03/05 22:42:57
File A: C:\windows\profiles\crossover\Desktop\My Mac Desktop\Human_Disease.flac
File B: C:\windows\profiles\crossover\Desktop\My Mac Desktop\Human_Disease.mp3
22:42:57 : Test started.
22:43:45 : 01/01 50.0%
22:44:03 : 02/02 25.0%
22:45:11 : 03/03 12.5%
22:45:31 : 03/04 31.3%
22:45:55 : 04/05 18.8%
22:46:19 : 05/06 10.9%
22:46:39 : 06/07 6.3%
22:46:53 : 07/08 3.5%
22:47:09 : 08/09 2.0%
22:47:32 : 09/10 1.1%
22:47:58 : 10/11 0.6%
22:48:00 : Test finished.
----------
Total: 10/11 (0.6%)
I would never have called it "very easy to ABX at V2". Actually it was quite a pain. Either LAME 3.98.2 is considerably better or you can still hear much higher frequencies than I can (I'm around the 30's). I can't really say what difference I heard. The background noise (from 0:01-0:03) just has another character somehow.
Your sample already contains very loud pre-ringing, probably due to heavily filtered hi-hats. I guess that's what the encoder is choking on, though both AAC encoders seem to handle it fine.
Your sample already contains very loud pre-ringing, probably due to heavily filtered hi-hats. I guess that's what the encoder is choking on, though both AAC encoders seem to handle it fine.
I also wounder if the hi-hats at the start casuses problems some encoders; but IMO i find this sample to have plenty of problems, such as the desortion later on the sample that Helix chokes on. Also on the full track there is tonal desortion problems with LAME 3.97 -V2 --vbr-new, i.e at around 3:54 - 4:01.
5.
Speaking of AAC encoders, I found out that Nero AAC still chokes at Linchpin at low bitrates.
Nero AAC 1.3.3.0 -q 0.40foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/07 16:02:18
File A: C:\Temp\Linchpin [Edit].flac
File B: C:\Temp\Linchpin q 0.40.mp4
16:02:18 : Test started.
16:02:30 : 01/01 50.0%
16:02:36 : 02/02 25.0%
16:02:47 : 03/03 12.5%
16:03:03 : 04/04 6.3%
16:03:12 : 05/05 3.1%
16:03:21 : 06/06 1.6%
16:03:30 : 07/07 0.8%
16:03:38 : 08/08 0.4%
16:03:46 : 09/09 0.2%
16:03:54 : 10/10 0.1%
16:04:01 : 11/11 0.0%
16:04:05 : 12/12 0.0%
16:04:11 : 13/13 0.0%
16:04:17 : 14/14 0.0%
16:04:28 : 15/15 0.0%
16:04:38 : 16/16 0.0%
16:04:44 : 17/17 0.0%
16:04:50 : 18/18 0.0%
16:05:03 : 19/19 0.0%
16:05:09 : 20/20 0.0%
16:05:10 : Test finished.
----------
Total: 20/20 (0.0%)
This was the sample that Helix choked on at the recent Mp3 listening test. Also Nero AAC seems to have same problem aswell, it sounds like something you would get from a Blade encoded mp3. The guitars warbles the looping drum beats sounds watery.
I re-uploaded the Linchpin sample, since some Nero AAC users might be interested to test it for themselves.
I can't hear the rain noise on my crappy laptop speakers but I sure do love Skinny Puppy
Yeah there awesome, shame i just only discovered them a few months ago when i got Minitsry's singles boxset that had a live cover of Smothered Hope.
The Human Disease sample was easy to ABX with LAME 3.98.2 -V2.
The problem is obvious in the first second of the sample with smearing/pre-echo, meaning no sharp attacks.
First, I started off with V6 just to get aware of the problem and then went higher to V4 and then finally to V2 which very lightly improved but obviously not enough to produce transparency.
Here is the result of the ABX test at V2:
foo_abx 1.3.3 report
foobar2000 v0.9.6
2009/03/10 19:53:30
File A: G:\backup\mmedia\Human_Disease.flac
File B: E:\Human_Disease.mp3
19:53:30 : Test started.
19:55:38 : 01/01 50.0%
19:55:52 : 02/02 25.0%
19:56:09 : 03/03 12.5%
19:56:57 : 04/04 6.3%
19:57:13 : 05/05 3.1%
19:57:17 : 06/06 1.6%
19:57:21 : 07/07 0.8%
19:57:31 : 08/08 0.4%
19:57:55 : 09/09 0.2%
19:58:19 : 10/10 0.1%
19:58:27 : 11/11 0.0%
19:58:37 : 12/12 0.0%
19:58:39 : Test finished.
----------
Total: 12/12 (0.0%)
I'll try to ABX this later at -320 just for the kicks.. I mean, for the attacks.
The Human Disease sample was easy to ABX with LAME 3.98.2 -V2.
The problem is obvious in the first second of the sample with smearing/pre-echo, meaning no sharp attacks.
First, I started off with V6 just to get aware of the problem and then went higher to V4 and then finally to V2 which very lightly improved but obviously not enough to produce transparency.
Glad to see someone else could pick out the horrid precho artifacts at the start of the sample. The sample is also easy to ABX at V0 on 3.97 and 3.98, for some reason at V0 with LAME 3.97 theres ringing, and at V0 with LAME 3.98 has no ringing but the precho artifact sounds a tad worse IMO and the extra 20kbps does not help at all. A sample like this makes me wonder why bother gaining a extra 50kbps avg bitrate per file for a little qaulity increase.
ABX logs:
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/10 21:19:14
File A: C:\Downloads\Human_Disease.flac
File B: C:\Temp\Human_Disease V0 LAME 3.97.mp3
21:19:14 : Test started.
21:19:45 : 01/01 50.0%
21:19:49 : 02/02 25.0%
21:19:56 : 03/03 12.5%
21:20:06 : 04/04 6.3%
21:20:15 : 05/05 3.1%
21:20:28 : 06/06 1.6%
21:20:38 : 07/07 0.8%
21:20:46 : 08/08 0.4%
21:20:52 : 09/09 0.2%
21:21:01 : 10/10 0.1%
21:21:09 : 11/11 0.0%
21:21:15 : 12/12 0.0%
21:21:16 : Test finished.
----------
Total: 12/12 (0.0%)
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/10 21:23:51
File A: C:\Downloads\Human_Disease.flac
File B: C:\Temp\Human_Disease V0 LAME 3.98.mp3
21:23:51 : Test started.
21:24:23 : 00/01 100.0%
21:24:27 : 01/02 75.0%
21:24:35 : 02/03 50.0%
21:24:45 : 03/04 31.3%
21:25:15 : 04/05 18.8%
21:25:31 : 05/06 10.9%
21:25:41 : 06/07 6.3%
21:25:51 : 07/08 3.5%
21:26:04 : 08/09 2.0%
21:26:10 : 09/10 1.1%
21:26:27 : 10/11 0.6%
21:26:53 : 11/12 0.3%
21:27:07 : Test finished.
----------
Total: 11/12 (0.3%)
Just as I had thought, this sample is also ABX-able with LAME 3.98.2 -b320 setting.
It was slightly better compared to V2 but not transparent again.
For me, what is ABX-able at V2, will usually also be ABX-able at -b320.
LAME 3.98.2 -b320:
foo_abx 1.3.3 report
foobar2000 v0.9.6
2009/03/11 06:07:49
File A: G:\backup\mmedia\Human_Disease.flac
File B: E:\Human_Disease.mp3
06:07:49 : Test started.
06:09:09 : 01/01 50.0%
06:09:19 : 02/02 25.0%
06:09:41 : 03/03 12.5%
06:10:04 : 04/04 6.3%
06:10:19 : 05/05 3.1%
06:11:01 : 06/06 1.6%
06:14:23 : 07/07 0.8%
06:18:23 : 08/08 0.4%
06:18:48 : 09/09 0.2%
06:19:29 : 10/10 0.1%
06:19:36 : 11/11 0.0%
06:19:49 : 12/12 0.0%
06:19:51 : Test finished.
----------
Total: 12/12 (0.0%)
P.S.: The name of the mp3 file is the same as before, but the file itself was overwritten with the -b320 setting.
Just as I had thought, this sample is also ABX-able with LAME 3.98.2 -b320 setting.
It was slightly better compared to V2 but not transparent again.
For me, what is ABX-able at V2, will usually also be ABX-able at -b320.
Someone else found it ABX-able at 320kbps CBR!, this must be a real problem sample. I could just managed to ABX the full track version of this sample around last December (http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=67840&view=findpost&p=603810) at 320, but yesterday i have failed to ABX it again at 320. The sample is pretty bad for anyone that is sensitve to smearing and precho artifacts.
/mnt, could you please let your golden ears try to identify a clearly ABX'able position in this encoding? I'm curious.
I have a theory about why you could ABX all those encodings so easily. If it is true, you should have massive problems with this encoding.
I have a theory about why you could ABX all those encodings so easily. If it is true, you should have massive problems with this encoding.
What encoding is this flac from? I know its lossy encoding with hardly no low-pass filter or something like that, i checked the md5sums from both orginal sample and your sample as wavs and they seem different.
I'll tell you after you have drawn your conclusion. Let's do this triple blind.
I'll tell you after you have drawn your conclusion. Let's do this triple blind.
Is the artifact at around 0:13, by any chance?
Anyway i performed the test about 15 mins ago:
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/15 19:45:13
File A: C:\Downloads\Human_Disease (rpp3po sample).flac
File B: C:\Downloads\Human_Disease.flac
19:45:13 : Test started.
19:45:49 : 01/01 50.0%
19:46:03 : 02/02 25.0%
19:46:13 : 03/03 12.5%
19:46:30 : 04/04 6.3%
19:46:42 : 05/05 3.1%
19:46:53 : 06/06 1.6%
19:47:13 : 07/07 0.8%
19:47:19 : 07/08 3.5%
19:47:24 : 08/09 2.0%
19:47:34 : 09/10 1.1%
19:47:46 : 10/11 0.6%
19:48:00 : 11/12 0.3%
19:48:07 : 12/13 0.2%
19:48:19 : 13/14 0.1%
19:48:37 : 14/15 0.0%
19:48:48 : 15/16 0.0%
19:48:55 : Test finished.
----------
Total: 15/16 (0.0%)
Sounds weak at around 0:13. And also at around 0:25 - 0:30, i thought i heard some kind of banging noise on my first test..
I'll tell you after you have drawn your conclusion. Let's do this triple blind.
Is the artifact at around 0:13, by any chance?
My guess would be that you were right on the spot for all other samples but this one is finally unABXable. But do your own tests.
I'll tell you after you have drawn your conclusion. Let's do this triple blind.
Is the artifact at around 0:13, by any chance?
My guess would be that you were right on the spot for all other samples but this one is finally unABXable. But do your own tests.
I did the test again at the start and it seems that the precho problem is gone at start:
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/15 20:28:14
File A: C:\Downloads\Human_Disease (rpp3po sample).flac
File B: C:\Downloads\Human_Disease.flac
20:28:14 : Test started.
20:28:45 : 00/01 100.0%
20:28:50 : 00/02 100.0%
20:28:55 : 00/03 100.0%
20:29:14 : 01/04 93.8%
20:29:21 : 02/05 81.3%
20:29:30 : 03/06 65.6%
20:29:42 : 03/07 77.3%
20:29:45 : 04/08 63.7%
20:29:58 : 04/09 74.6%
20:30:05 : 05/10 62.3%
20:30:10 : 06/11 50.0%
20:30:16 : 06/12 61.3%
20:30:24 : 06/13 70.9%
20:30:29 : 07/14 60.5%
20:30:34 : 07/15 69.6%
20:30:43 : 07/16 77.3%
20:30:51 : 08/17 68.5%
20:31:00 : 09/18 59.3%
20:31:04 : 09/19 67.6%
20:31:10 : 10/20 58.8%
20:31:11 : Test finished.
----------
Total: 10/20 (58.8%)
Wow, how can you hear that? I wouldn't have expected this. I thought your key to success is solely a very high cut-off in combination with the shabby treatment of most codec presets of the >16k Hz range. I have found out that even the highest quality setting of QT's true VBR still cuts at 18k, which I don't appreciate. People being able to hear higher frequencies should at least be able to switch this off. This last sample was iTunes plus (256 kbit/s constrained VBR), which straightforwardly preserves anything until over 20khz.
So I have made some measurements to find out why in hell you can still hear a difference. There are several indications, that you are not cheating:
1. You have identified the same problem spots as the encoder shows with other settings without knowing it (the only exception would be that you have compared checksums against an iTunes plus encode - reading today's discussions could have put you on the track).
2. Metrical analysis of the sound material shows other hot spots and you haven't just reported those.
For further investigation I generated the delta file between the lossless and lossy version. For other tracks those have often been almost empty (except for high frequency bleeps and maybe several hotspots). In the case of your sample the delta file is completely filled with loud, audible material from the beginning. You instantly recognize the source track over the whole range. So this really is quite a problem sample. Maybe you should contact the developers.
Wow, how can you hear that? I wouldn't have expected this. I thought your key to success is solely a very high cut-off in combination with the shabby treatment of most codec presets of the >16k Hz range. I have found out that even the highest quality setting of QT's true VBR still cuts at 18k, which I don't appreciate. People being able to hear higher frequencies should at least be able to switch this off. This last sample was iTunes plus (256 kbit/s constrained VBR), which straightforwardly preserves anything until over 20khz.
I can not hear anything over 18.5k - 19k Hz, am just sensitive to precho and warbling artifacts and others aswell. Also the first test of your sample was sorta luck and being very picky at around 0.13.
Are you sure about your inability to ABX Nero 1.3.3.0 -q 0.55?
If yes, that would be an impressive victory for Nero. Well, only one sample - but a very tough one.
Are you sure about your inability to ABX Nero 1.3.3.0 -q 0.55?
If yes, that would be an impressive victory for Nero. Well, only one sample - but a very tough one.
I have passed a few ABX test with Nero AAC at 0.55 a number of times, but at q 0.53 it takes alot more effort to do so just like LAME -V 1. Am sure this sample could be ABXable at 0.55 if i really try harder; but that would make me more sensitive to precho. Anyway Nero AAC chokes just as bad as Helix mp3 on that Linchpin sample i uploaded.
A second result would be that iTunes Plus (QT 256kbit/s VBR constrained) is not (considerably) better than Quicktime True VBR, albeit using more than 30% more disk space, with problematic source material. Not forgetting that we are talking about a very high quality level, which most people cannot differentiate at all and also you only with difficulties.
A second result would be that iTunes Plus (QT 256kbit/s VBR constrained) is not (considerably) better than Quicktime True VBR, albeit using more than 30% more disk space, with problematic source material. Not forgetting that we are talking about a very high quality level, which most people cannot differentiate at all and also you only with difficulties.
I found the iTunes and Quicktime encodes did a good job making the precho problem transparent; but they seem to struggle with the disortion noises later on the sample.
You're probably through for today. Attached you'll find Linchpin QT encodes, in case you should be interested later.
You're probably through for today. Attached you'll find Linchpin QT encodes, in case you should be interested later.
I'll try them tommorrow, am been meaning to test this sample out with iTunes AAC weeks ago.
The encoding mode I made those with, is only available in iTunes since 8.1, which is only a couple of days old. Maybe it's better, maybe not.
The encoding mode I made those with, is only available in iTunes since 8.1, which is only a couple of days old. Maybe it's better, maybe not.
Likely be the same; but there has been known regression issues with iTunes + Quicktime in the past. Such as the thundershock (or whatever it was called) sample.
I have followed Quicktime AAC for quite some time, because I was always out for a native Apple solution. Quicktime AAC wasn't really all that great and I used Nero through Wine for a long time. And you are right it even became worse over time. This changed at some point, I think it was Quicktime 7.2. Since then quality is improving again. Especially Quicktime 7.6 was a huge step up for higher bitrates.
I have found out that even the highest quality setting of QT's true VBR still cuts at 18k, which I don't appreciate. People being able to hear higher frequencies should at least be able to switch this off. This last sample was iTunes plus (256 kbit/s constrained VBR), which straightforwardly preserves anything until over 20khz.
Hmm.. I just checked the q127 file and the cutoff seems to be around 19khz.
Hmm.. I just checked the q127 file and the cutoff seems to be around 19khz.
I can confirm that. At 19khz is a steep cutoff, not at 18khz and not only for this file. I have checked the track that I had based my former statement on again and it's also 19khz. Must have confounded the columns. True VBR seems fine then; I don't consider 19khz problematic. LAME's 16khz (if you couldn't switch it off for young ears) yes, but 18khz was already a borderline case and 19khz shouldn't really make a difference.
@ rpp3po
I did a test with those iTunes AAC files, and it seems that Nero AAC does has some issues with that sample.
iTunesfoo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/16 21:07:20
File A: C:\Temp\Linchpin [Edit].flac
File B: C:\Downloads\Linchpin__QtTrueVBR127.m4a
21:07:20 : Test started.
21:08:04 : 01/01 50.0%
21:08:47 : 02/02 25.0%
21:09:03 : 02/03 50.0%
21:09:08 : 03/04 31.3%
21:09:34 : 03/05 50.0%
21:09:47 : 04/06 34.4%
21:09:58 : 05/07 22.7%
21:10:07 : 06/08 14.5%
21:10:19 : 06/09 25.4%
21:10:31 : 06/10 37.7%
21:10:36 : 07/11 27.4%
21:10:44 : 08/12 19.4%
21:10:54 : 09/13 13.3%
21:11:05 : 09/14 21.2%
21:11:10 : 09/15 30.4%
21:11:20 : 10/16 22.7%
21:11:29 : 11/17 16.6%
21:11:45 : 12/18 11.9%
21:12:03 : 13/19 8.4%
21:12:05 : Test finished.
----------
Total: 13/19 (8.4%)
Am not sure about this encode, it was hard to ABX.
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/16 21:12:45
File A: C:\Temp\Linchpin [Edit].flac
File B: C:\Downloads\Linchpin__iTunesPlus.m4a
21:12:45 : Test started.
21:13:41 : 00/01 100.0%
21:13:49 : 00/02 100.0%
21:14:28 : 00/03 100.0%
21:14:48 : 00/04 100.0%
21:15:04 : 00/05 100.0%
21:15:10 : 00/06 100.0%
21:15:14 : 00/07 100.0%
21:15:18 : 01/08 99.6%
21:15:26 : 02/09 98.0%
21:15:43 : 03/10 94.5%
21:15:57 : 04/11 88.7%
21:16:03 : 04/12 92.7%
21:16:12 : 04/13 95.4%
21:16:16 : 04/14 97.1%
21:16:21 : 04/15 98.2%
21:16:22 : Test finished.
----------
Total: 4/15 (98.2%)
Seems transparent to me.
Nero AAC 1.3.3.0q 0.50
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/16 21:17:58
File A: C:\Temp\Linchpin [Edit].flac
File B: C:\Temp\Linchpin.mp4
21:17:58 : Test started.
21:18:22 : 01/01 50.0%
21:18:34 : 02/02 25.0%
21:18:47 : 03/03 12.5%
21:19:04 : 04/04 6.3%
21:19:18 : 05/05 3.1%
21:19:31 : 06/06 1.6%
21:19:43 : 07/07 0.8%
21:19:49 : 08/08 0.4%
21:19:57 : 09/09 0.2%
21:20:07 : 10/10 0.1%
21:20:23 : 11/11 0.0%
21:20:38 : 12/12 0.0%
21:20:39 : Test finished.
----------
Total: 12/12 (0.0%)
Better then q 0.40, but still full of smearing and warbling.
q 0.55
foo_abx 1.3.3 report
foobar2000 v0.9.6.3
2009/03/16 21:21:37
File A: C:\Temp\Linchpin [Edit].flac
File B: C:\Temp\Linchpin q 0.55.m4a
21:21:37 : Test started.
21:21:53 : 01/01 50.0%
21:22:06 : 02/02 25.0%
21:22:19 : 02/03 50.0%
21:22:25 : 03/04 31.3%
21:22:35 : 04/05 18.8%
21:22:39 : 05/06 10.9%
21:22:51 : 06/07 6.3%
21:23:09 : 06/08 14.5%
21:23:14 : 07/09 9.0%
21:23:24 : 07/10 17.2%
21:23:28 : 08/11 11.3%
21:23:32 : 09/12 7.3%
21:23:40 : 10/13 4.6%
21:23:45 : 11/14 2.9%
21:23:54 : 12/15 1.8%
21:24:06 : 13/16 1.1%
21:24:14 : 14/17 0.6%
21:24:18 : 15/18 0.4%
21:24:23 : 16/19 0.2%
21:24:30 : 17/20 0.1%
21:24:45 : 18/21 0.1%
21:24:53 : 19/22 0.0%
21:24:55 : Test finished.
----------
Total: 19/22 (0.0%)
An improvement over q 0.50, smearing is gone but theres warbling on the guitars. Dammit that track is not transparent to me at q 0.55 anymore
Thank you /mnt, your ears are golden! That's very interesting input. And all this despite your heavy metal consumption... Or maybe exactly that made you so sensitive for compression artifacts, because it is so hard to compress properly.
I have repeated the ABX tests. I fail at both QT encodes, but Nero VBR at q .5 is really very easy as you say, even to my ears. But I couldn't really get .55 anymore and instantly failed. Then I somehow got a hold of some wobbling, but
very faintly. After some tries I lost it again and was already feeling some fatigue. If I had continued after No. 23 the results would have got worse and worse. Anyway, the Nero devs should have a look at this. q 5 is really a good general setting for most cases. It would be even better without slips like that.
foo_abx 1.3.3 report
foobar2000 v0.9.6.2
2009/03/16 23:24:33
File A: C:\windows\profiles\crossover\Desktop\My Mac Desktop\Linchpin__Edit_.flac
File B: C:\windows\profiles\crossover\Desktop\My Mac Desktop\Linchpin__neroq5.m4a
23:24:33 : Test started.
23:25:22 : 01/01 50.0%
23:25:36 : 02/02 25.0%
23:25:50 : 03/03 12.5%
23:26:13 : 04/04 6.3%
23:26:29 : 05/05 3.1%
23:26:43 : 06/06 1.6%
23:27:03 : 07/07 0.8%
23:27:20 : 08/08 0.4%
23:27:23 : Test finished.
----------
Total: 8/8 (0.4%)
foo_abx 1.3.3 report
foobar2000 v0.9.6.2
2009/03/16 23:29:22
File A: C:\windows\profiles\crossover\Desktop\My Mac Desktop\Linchpin__Edit_.flac
File B: C:\windows\profiles\crossover\Desktop\My Mac Desktop\Linchpin__neroq55.m4a
23:29:22 : Test started.
23:29:30 : 00/01 100.0%
23:29:46 : 00/02 100.0%
23:30:03 : 01/03 87.5%
23:30:38 : 02/04 68.8%
23:31:02 : 03/05 50.0%
23:31:25 : 04/06 34.4%
23:31:41 : 04/07 50.0%
23:32:05 : 05/08 36.3%
23:32:43 : 06/09 25.4%
23:33:09 : 07/10 17.2%
23:33:41 : 08/11 11.3%
23:34:01 : 09/12 7.3%
23:34:29 : 10/13 4.6%
23:34:42 : 11/14 2.9%
23:35:01 : 12/15 1.8%
23:35:14 : 13/16 1.1%
23:35:27 : 13/17 2.5%
23:35:50 : 13/18 4.8%
23:36:16 : 13/19 8.4%
23:36:33 : 14/20 5.8%
23:36:52 : 14/21 9.5%
23:37:19 : 15/22 6.7%
23:37:45 : 15/23 10.5%
23:37:50 : Test finished.
----------
Total: 15/23 (10.5%)
Thank you /mnt, your ears are golden! That's very interesting input. And all this despite your heavy metal consumption... Or maybe exactly that made you so sensitive for compression artifacts, because it is so hard to compress properly.
I have repeated the ABX tests. I fail at both QT encodes, but Nero VBR at q .5 is really very easy as you say, even to my ears. But I couldn't really get .55 anymore and instantly failed. Then I somehow got a hold of some wobbling, but very faintly. After some tries I lost it again and was already feeling some fatigue. If I had continued after No. 23 the results would have got worse and worse. Anyway, the Nero devs should have a look at this. q 5 is really a good general setting for most cases. It would be even better without slips like that.
Thanks for testing the sample out and also proved it was easy to ABX at q 0.50, I posted a few ABX logs on the Nero AAC 1.3.3.0 topic.
Hi all,
thought I should recycle this thread. I found a single-tone sample from an 80's Roland MKS-50 digitally controlled analog synth which I recorded back in 2004 in 16-bit/96-kHz. I converted this to 16-bit/44.1-kHz this morning and - just for fun - tried a foobar ABX vs. LAME -V3. This is what I got.
C:\>lame -V3 B86_WhatThe_Chorus.wav
LAME 3.98.2 32bits (http://www.mp3dev.org/)
CPU features: MMX (ASM used), 3DNow! (ASM used), SSE (ASM used), SSE2
Using polyphase lowpass filter, transition band: 17960 Hz - 18494 Hz
Encoding B86_WhatThe_Chorus.wav to B86_WhatThe_Chorus.wav.mp3
Encoding as 44.1 kHz j-stereo MPEG-1 Layer III VBR(q=3)
Frame | CPU time/estim | REAL time/estim | play/CPU | ETA
1074/1074 (100%)| 0:01/ 0:01| 0:01/ 0:01| 15.886x| 0:00
32 [ 16] **
40 [ 0]
48 [ 0]
56 [ 1] %
64 [ 0]
80 [ 0]
96 [ 25] %%*
112 [ 95] %%%%%%%%***
128 [ 105] %%%%%%******
160 [ 74] %********
192 [ 17] %*
224 [ 82] **********
256 [ 41] *****
320 [ 618] %*******************************************************************
-------------------------------------------------------------------------------
kbps LR MS % long switch short %
250.3 13.0 87.0 25.4 9.6 65.0
Writing LAME Tag...done
ReplayGain: -6.6dB
_______________________________________________________________________________
foo_abx 1.3.4 report
foobar2000 v0.9.6.5
2009/05/16 13:41:55
File A: C:\B86_WhatThe_Chorus.wav
File B: C:\B86_WhatThe_Chorus_mp3.wav
13:41:55 : Test started.
13:42:49 : 01/01 50.0%
13:44:22 : 02/02 25.0%
13:45:04 : 03/03 12.5%
13:45:39 : 04/04 6.3%
13:46:31 : 04/05 18.8%
13:47:21 : 05/06 10.9%
13:48:14 : 06/07 6.3%
13:49:16 : 07/08 3.5%
13:49:24 : Test finished.
----------
Total: 7/8 (3.5%)
The MP3 (decoded to 16/44 WAV and delay-compensated) sounds excellent, but apparently is not transparent. There are two tones in the excerpt. I listened to each tone 4 times. Audio setup: Sennheiser HD-590 directly connected to a Terratec Phase 22 via Y cable.
I also uploaded the 16/96 version since it contains a lot of content above 22 kHz. I might try to ABX the 16/44 FLAC against the 16/96 FLAC soon.
Chris
The files
Warning: LOUD![attachment=5105:B86_What...us_44kHz.flac]
[attachment=5106:B86_What...us_96kHz.flac]
With Nero AAC 1.3.3.0 at -q 0.55 I managed to ABX Human Disease 13/13 (pre-echo and stereo image where there is none in the original) and Linchpin 8/8 (smeared hihat attacks).
EDIT: Attached reports
foo_abx 1.3.4 report
foobar2000 v0.9.6.9
2009/09/17 17:11:41
File A: C:\temp\_ - _ - Human_Disease.mp4
File B: C:\temp\Human_Disease.flac
17:11:41 : Test started.
17:12:03 : Trial reset.
17:14:39 : 01/01 50.0%
17:14:56 : 02/02 25.0%
17:15:38 : 03/03 12.5%
17:16:04 : 04/04 6.3%
17:16:24 : 05/05 3.1%
17:16:34 : 06/06 1.6%
17:16:42 : 07/07 0.8%
17:16:50 : 08/08 0.4%
17:17:01 : 09/09 0.2%
17:17:13 : 10/10 0.1%
17:17:26 : 11/11 0.0%
17:17:37 : 12/12 0.0%
17:18:00 : 13/13 0.0%
17:18:08 : Test finished.
----------
Total: 13/13 (0.0%)
foo_abx 1.3.4 report
foobar2000 v0.9.6.9
2009/09/17 17:19:29
File A: C:\temp\Linchpin__Edit_.flac
File B: C:\temp\05 - Fear Factory - Linchpin.mp4
17:19:29 : Test started.
17:19:49 : 01/01 50.0%
17:20:02 : 02/02 25.0%
17:20:40 : 03/03 12.5%
17:21:42 : 04/04 6.3%
17:21:51 : 05/05 3.1%
17:22:03 : 06/06 1.6%
17:22:16 : 07/07 0.8%
17:22:31 : 08/08 0.4%
17:23:18 : Test finished.
----------
Total: 8/8 (0.4%)
With Nero AAC 1.3.3.0 at -q 0.55 I managed to ABX Human Disease 13/13 (pre-echo and stereo image where there is none in the original) and Linchpin 8/8 (smeared hihat attacks).
Sounds like the same problems i spotted aswell. I didn't noticed the stereo image artifact on Human Disease though, just the pre-echo that makes it sound like a water sprinkler.
This sample is also ABXable with QuickTime True VBR at 60:
foo_abx 1.3.4 report
foobar2000 v1.0
2010/01/18 03:17:56
File A: C:\Temp\Human_Disease.flac
File B: C:\Temp\Human_Disease QT TVBR 60.m4a
03:17:56 : Test started.
03:18:14 : 01/01 50.0%
03:18:27 : 02/02 25.0%
03:18:39 : 03/03 12.5%
03:18:56 : 04/04 6.3%
03:19:14 : 05/05 3.1%
03:19:24 : 06/06 1.6%
03:19:47 : 07/07 0.8%
03:20:16 : 08/08 0.4%
03:20:29 : 09/09 0.2%
03:20:55 : 10/10 0.1%
03:20:58 : Test finished.
----------
Total: 10/10 (0.1%)
It has pre-echo and the disortions past 0:15 sounds messy.