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: Various pictures from Arny's posts (Read 89368 times) previous topic - next topic
0 Members and 4 Guests are viewing this topic.

Various pictures from Arny's posts

Reply #50
Visual comparison of 7-14-2014 versions of the keys jangling files

[attachment=8258:keys_jan...sus_1644.png]

Please notice approximate 150 microsecond timing difference

Various pictures from Arny's posts

Reply #51
For reference, the clearly audible 10 millisecond timing error between Mosaic A2 and Mosiac B2 from the AVS-AIX sample rate comparison files

[attachment=8260:Mosiac_a...ng_error.png]

Amd here is  my evidence confirming that the difference is audible:

Code: [Select]
foo_abx 2.0 beta 4 report
foobar2000 v1.3.5
2015-04-08 19:24:01

File A: Mosaic_A2.wav
SHA1: 8ff31de028091c0daadb36c36bb51cbf88548e13
File B: Mosaic_B2.wav
SHA1: b899f3be763f92c642f7f6475f8cf3efad3023e7

Output:
DS : Primary Sound Driver

19:24:01 : Test started.
19:31:27 : 01/01
19:31:44 : 02/02
19:32:12 : 03/03
19:32:36 : 04/04
19:33:36 : 05/05
19:33:48 : 06/06
19:34:11 : 07/07
19:34:37 : 08/08
19:34:48 : 09/09
19:35:34 : 10/10
19:36:23 : 11/11
19:36:45 : 12/12
19:36:59 : 13/13
19:37:32 : 14/14
19:37:41 : 15/15
19:37:50 : 16/16
19:37:50 : Test finished.

----------
Total: 16/16
Probability that you were guessing: 0.0%

-- signature --
08f592c6e0dc1934c5435c20c95509a77000cd15


Methodology I used was to isolate a segment in the 3:24-3:34 range where the tell was most audible, and just switch back and forth between A and X and A and X until I heard the telltale echo, and if I failed there, or wished to confirm, did the same thing with B and X.

Various pictures from Arny's posts

Reply #52
What output device do you have that doesn't have IM, Wombat?


There can be no such thing. Everything has nonlinear distortion, its just that for a lot of things like pure copper wire, its down quite a few dB.

The best I've seen is my AP 24192 that has all forms of distortion about 100 dB down over the 3 sets of test frequencies.

Various pictures from Arny's posts

Reply #53
I don't know if i have much IM. I only know i don't hear it for these test samples at the loudnes i abx the keys.
The interesting thing is to what deggree IM is underrated as being a problem in all these comparisons.
We already wondered if this Meridian/AES test setup only was the most complicated setup ever to prove IM and not some filters.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Various pictures from Arny's posts

Reply #54
I don't know if i have much IM. I only know i don't hear it for these test samples at the loudness I abx the keys.


That is the desired outcome - both files sound the same through their test tone region.  They should sound the same and they should be silent.

Quote
The interesting thing is to what degree IM is underrated as being a problem in all these comparisons.


The purpose of this test adjunct is to let people find that out for themselves in exactly the same context as the rest of the test.

Quote
We already wondered if this Meridian/AES test setup only was the most complicated setup ever to prove IM and not some filters.


Good question!

IME the most common source of audible IM is clipping.

Various pictures from Arny's posts

Reply #55
Meridian's CD bashing  "typical" filter audibility paper made no claims to have tested for:

A) IM at the ear [could have been from their wacky, self amplified speakers, keep in mind, perhaps even in the transducers themselves let alone their electronics],

B) time synch accuracy [although refusing the listeners the ability to switch on the fly at their own will means this is less important, but not completely unimportant], and

C) precise level matching in the audible band

Even if they did use some level matching procedure, I bet they would have used something with, for instance, a simple K-weighted filter, like I believe Replaygain uses, which might very well be sensitive to ultasonics and therefore mis-matches the level in the audible, <20kHz range in order to be "correct" for the wideband average level.

Various pictures from Arny's posts

Reply #56
Even if they did use some level matching procedure, I bet they would have used something with, for instance, a simple K-weighted filter, like I believe Replaygain uses, which might very well be sensitive to ultasonics and therefore mis-matches the level in the audible, <20kHz range in order to be "correct" for the wideband average level.

Whoa, this seems like a big problem to me. 

Just threw together some samples in audacity of a 1kHz tone with amplitude 0.01, and in a second file I added a loud (0.8 amplitude) 24kHz tone:

[attachment=8261:0.01_amp...khz_tone.flac]
[attachment=8262:0.01_amp...khz_tone.flac] (warning: contains near full-scale ultrasonic tone!!!)

Replaygain (in foobar2k) reports wildly different track gains (+22 dB and -19 dB).
loudness-scanner (https://github.com/jiixyj/loudness-scanner), a command-line tool implementing EBU R128, reports -40.0 LUFS and 1.4 LUFS.
yet, they sound about the same volume, of course, because we can't hear the 24kHz tone.

I can see why you'd want a loudness measurement to alert you to loud ultrasonics in some contexts, but it seems to me that for level matching tracks in an ABX test, we really need a loudness measurement that takes into account the decreasing sensitivity of our hearing above ~15kHz, and totally ignores ultrasonics. i.e. respecting the right edge of these equal loudness curves: http://en.wikipedia.org/wiki/Loudness#/med...ile:Lindos1.svg

Various pictures from Arny's posts

Reply #57
That wasn't such a big problem with the original ReplayGain

Though as I keep telling people, for near-transparent double-blind testing of objectively different audio signals, you really shouldn't use ReplayGain (any version) to match levels. It's not designed for that. Find some other way. e.g. knowing what the difference is in the audible band as part of the test design, and correcting for it.

Cheers,
David.

Various pictures from Arny's posts

Reply #58
Luckily the latest version of foobar ABX shows in the summarized test results when ami r a user engages Replaygain [with its sensitivity to ultrasonics] in order to cheat, due to this very vulnerability I mentioned. Unfortunately this critical info was not part of the summarized report back when Arny first released his jangling keys test files in the AVS AIX records thread nor when the AIX records files themselves were released [making the posted test results of others rather questionable for all of these various ultrasonic tests, back then].

Various pictures from Arny's posts

Reply #59
That wasn't such a big problem with the original ReplayGain

Though as I keep telling people, for near-transparent double-blind testing of objectively different audio signals, you really shouldn't use ReplayGain (any version) to match levels. It's not designed for that. Find some other way. e.g. knowing what the difference is in the audible band as part of the test design, and correcting for it.


When I'm preparing test files for ABXing, right up front I tack a few test tones in the normal audible range (20 Hz, 1 KHz, 16 KHz) onto the front or back of the file(s), process it, and then check  to see if the test tones are still at the desired levels, signal purity, etc.

Nothing new about this - good recording, mixing and mastering engineers have been doing this for decades - nearly half a century.  Doing this was particularly important when people were using Dolby Noise Reduction.

Various pictures from Arny's posts

Reply #60
The following is a 22.0 KHz tone burst downsampled to 4416 CEP 2.1 Q=999 and upsampled back to 2496

[attachment=8263:2496_22_...16_Q_999.png]

The following is a 22.0 KHz tone burst downsampled to 4416 CEP 2.1 Q=150 and upsampled back to 2496

[attachment=8264:2496_22_...16_Q_150.png]

Various pictures from Arny's posts

Reply #61
Some files for interchannel delay listening tests

[attachment=8278:impulses_2klp_4416.zip]

There is a larger collection of files, these are the highlights - reference. 1 sample, 9 samples, 27 samples

(1) A 0.45 second 24/192 file with a unit impulse in the middle was created
(2) It was appended 3 times to create a file with 4 impulses
(3) A number of copies were created with various delays added to the Left channel. 1,2,3,4,5,7,9,11,15,19,27 samples @ 192 kHz. Units of about 5 uSec
(4) All files were low pass filtered @ 2 KHz with a 10th order minimum phase low pass filter
(5) All files were down sampled to 4416
(6) all files were normalized to 90% (-1 dB FS)


Various pictures from Arny's posts

Reply #62
Thanks. I'm going out now but will examine them more closely later.

edit: Preliminary ABX now posted in other thread and deleted from this post.

Various pictures from Arny's posts

Reply #63
Some files for interchannel delay listening tests

Attached File 

[attachment=8280:impulses...klp_4416.zip]

This is the upudated/corrected version.  The 27 step delay file is designed to be very easy and give you an idea of what to listen to.

Test for 2 and 4 step delays were added to put in some more tests between doable and impossible.

1 sample delay = 5.2 uSec
2 sample delay = 10.4 uSec
4 sample delay = 20.8 uSec
9 sample delay  = 46.85 uSec  - This is roughly the result that David L. Clark reported.
27 sample delay  = 140.6  uSec - this is a confidence builder - you should be able to complete it easily and accurately. It is also a test of the suitability of your test environment.  Do not proceed to the shorter delays until you can do well with this file.

By the time you work your way down to the 1 step delay file, expect to sweat a little! ;-)

There is a larger collection of files, these are the highlights - reference. 1 sample delay, 2 samples delay, 4 samples delay,  9 samples delay, 27 samples delay.

Please start out with the 27 sample delay and work your way down.

This is how the files were created in CEP 2.1

(1) A 0.45 second 24/192 file with a unit impulse in the middle was created
(2) It was appended 3 times to create a file with 4 impulses
(3) A number of copies were created with various delays added to the Left channel. 1,2,3,4,5,7,9,11,15,19,27 samples @ 192 kHz. Units of about 5 uSec
(4) All files were low pass filtered @ 2 KHz with a 10th order minimum phase low pass filter
(5) All files were down sampled to 4416
(6) all files were normalized to 90% (-1 dB FS)

Various pictures from Arny's posts

Reply #64
Attached files generated by:

Code: [Select]
for d in 0 1; do sox -V -n delay-${d}.flac synth 1s sq pad 100s 379s sinc -a 82 -t 4k -M 4k repeat 399 channels 2 delay ${d}s norm -1 trim 0 4; done



Various pictures from Arny's posts

Reply #65
Attached files generated by:

Code: [Select]
for d in 0 1; do sox -V -n delay-${d}.flac synth 1s sq pad 100s 379s sinc -a 82 -t 4k -M 4k repeat 399 channels 2 delay ${d}s norm -1 trim 0 4; done



My analysis is that the file with the delay adds a 1 sample (22 uSec) delay to the left channel where the pulse rate is 100 Hz.

Can this methodology produce delays that are other than 1 sample?

Can this methodology produce delays that are less than 1 sample?

What are the typical ABX results?

Various pictures from Arny's posts

Reply #66
Added test file with 3 sample shift.  This file includes reference file and 5 previous test files;

[attachment=8289:impulses...klp_4416.zip]

This is the updated/corrected version. The 27 step delay file is designed to be very easy and give you an idea of what to listen to.

Test for 2 and 4 step delays were added to put in some more tests between doable and impossible.

1 sample delay = 5.2 uSec
2 sample delay = 10.4 uSec
3 sample delay = 15.6 uSec
4 sample delay = 20.8 uSec
9 sample delay = 46.85 uSec - This is roughly the result that David L. Clark reported.
27 sample delay = 140.6 uSec - this is a confidence builder - you should be able to complete it easily and accurately. It is also a test of the suitability of your test environment. Do not proceed to the shorter delays until you can do well with this file.

By the time you work your way down to the 2 sample delay file, expect to sweat a little! ;-)

There is a larger collection of files, these are the highlights - reference. 1 sample delay, 2 samples delay, 4 samples delay, 9 samples delay, 27 samples delay.

Please start out with the 27 sample delay and work your way down.

This is how the files were created in CEP 2.1

(1) A 0.45 second 24/192 file with a unit impulse in the middle was created
(2) It was appended 3 times to create a file with 4 impulses
(3) A number of copies were created with various delays added to the Left channel. 1,2,3,4,5,7,9,11,15,19,27 samples @ 192 kHz. Units of about 5 uSec
(4) All files were low pass filtered @ 2 KHz with a 10th order minimum phase low pass filter
(5) All files were down sampled to 4416
(6) all files were normalized to 90% (-1 dB FS)


 

Various pictures from Arny's posts

Reply #68
Attached files generated by: 
Code: [Select]
for d in 0 1; do sox -V -n delay-${d}.flac synth 1s sq pad 100s 379s sinc -a 82 -t 4k -M 4k repeat 399 channels 2 delay ${d}s norm -1 trim 0 4; done


 The sound is irritating, and seems right at my threshold based on Arny's test, so I didn't ABX these until just now. First attempt:

Code: [Select]

foo_abx 2.0 report
foobar2000 v1.3.3
2015-05-03 11:03:33

File A: delay-0.flac
SHA1: bf47cf35f523323d48e21e0b78b4627663c6021e
File B: delay-1.flac
SHA1: 8f34e5bc23ad74724b86162f768a1391939b7603

Output:
DS : Primary Sound Driver
Crossfading: NO

11:03:33 : Test started.
11:04:39 : 01/01
11:04:55 : 02/02
11:05:08 : 02/03
11:05:32 : 03/04
11:06:21 : 04/05
11:06:36 : 04/06
11:06:49 : 05/07
11:07:06 : 06/08
11:07:06 : Test finished.

 ----------
Total: 6/8
Probability that you were guessing: 14.5%

 -- signature --
7a608ad88099aba9ecc6db3265ab8fda761576bb


Is this where I should be posting my results or some other thread?

I'll try again, I'm pretty sure I can ace this, but I have to wait for my refrigerator compressor to turn off so I can have dead silence, to concentrate.

edit to add: Second test had similar results. Third attempt here below. SCORE!

Code: [Select]

foo_abx 2.0 report
foobar2000 v1.3.3
2015-05-03 11:24:55

File A: delay-0.flac
SHA1: bf47cf35f523323d48e21e0b78b4627663c6021e
File B: delay-1.flac
SHA1: 8f34e5bc23ad74724b86162f768a1391939b7603

Output:
DS : Primary Sound Driver
Crossfading: NO

11:24:55 : Test started.
11:26:30 : 01/01
11:26:51 : 02/02
11:29:49 : 03/03
11:30:11 : 04/04
11:30:56 : 05/05
11:31:16 : 06/06
11:31:40 : 07/07
11:31:57 : 08/08
11:32:15 : 09/09
11:32:36 : 10/10
11:32:36 : Test finished.

 ----------
Total: 10/10
Probability that you were guessing: 0.1%

 -- signature --
20caca8e62ab90c17b000f86cf8d8c607c9dedfe


Various pictures from Arny's posts

Reply #69
P.S. I never in a million years would have thought that I could discern just a one sample shift. It seems too small. I'm suspicious there could be other elements such as artifacts I'm keying on. This is right on the cusp of my detection capabilities so I can't even say with 100% certainty what exactly it is I'm hearing that changes. Is it level? is it balance? is it tonality? Is it directional cues? Hmm, don't know. I'd say "directional cues" but maybe that's just from my observer bias kicking in?

When you guys make these files I'm thinking you shouldn't take a master file "A" and then present a secondary, delayed file version of it for "B", but unstead should take a master file, delay it by say 1 sample, call that "A", and then take the same master file and delay it 2 samples and call that one "B". This way both A and B have undergone processing, which may be introducing artifacts, but at least it will be the same for both A and B.

Various pictures from Arny's posts

Reply #70
P.S. I never in a million years would have thought that I could discern just a one sample shift. It seems too small. I'm suspicious there could be other elements such as artifacts I'm keying on. This is right on the cusp of my detection capabilities so I can't even say with 100% certainty what exactly it is I'm hearing that changes. Is it level? is it balance? is it tonality? Is it directional cues? Hmm, don't know. I'd say "directional cues" but maybe that's just from my observer bias kicking in?


That's one of the beauties of doing DBTs with enough trials. When you hear something, even if it is too small to characterize, the statistics make it clear that you are hearing a diffference, and the controls ensure that you are hearing something.

I don't think that people who can't trust DBTs ever get into them enough to have this experience, and thus they miss out on a really convincing experience. I don't recall when we started experiencing this, but it was clearly happening 10 years after the first ABX tests.

Quote
When you guys make these files I'm thinking you shouldn't take a master file "A" and then present a secondary, delayed file version of it for "B", but unstead should take a master file, delay it by say 1 sample, call that "A", and then take the same master file and delay it 2 samples and call that one "B". This way both A and B have undergone processing, which may be introducing artifacts, but at least it will be the same for both A and B.


You already have some files with differential delays of that kind to listen to.

However, these particular files made by me are made by a particularly reliable procedure involving adding samples of silence to the front of one channel and not the other. The whole file is not processed by a delaying process, and the undelayed channel isn't processed by any delaying mechanism at all.

Various pictures from Arny's posts

Reply #71
Arny, please post a 5 sample delay version of your recent tests. I will then attempt to ABX it not against the reference file but instead against the 1 sample delay version and then will know if what I am hearing is truly just a 4 sample delay difference or if your processing is introducing artifacts in the first test I took, the one where I heard a difference at the 4 sample, but not 3 sample, level of delay. Thanks.

[I also can attempt to compare 5 sample delay against 2 sample delay to see if for some unforeseen reason I can detect only a 3 sample difference that I couldn't before.]


Various pictures from Arny's posts

Reply #72
Arny, please post a 5 sample delay version of your recent tests. I will then attempt to ABX it not against the reference file but instead against the 1 sample delay version and then will know if what I am hearing is truly just a 4 sample delay difference or if your processing is introducing artifacts in the first test I took, the one where I heard a difference at the 4 sample, but not 3 sample, level of delay. Thanks.

[I also can attempt to compare 5 sample delay against 2 sample delay to see if for some unforeseen reason I can detect only a 3 sample difference that I couldn't before.]


Since I'm on the road and don't have the original 24/192 impulses on my laptop, I'd like to put that off until I return home.

Yes, I could recreate them, but I might screw that up in some subtle way. So if you would be so kind as to wait a week... Pretty Please?;-)

While I'm cleaning that project up, you said something about background noise. I think I can make that go away. Necessary to fix?

Also, there might be a place in life for the whole file sequence to be recreated, only with a 2K High Pass filter in place of the 2K low pass filter. That would be closer to JJ's suggestion. Both need to be tested, and the compare & contrast analysis might be interesting.

Comments, anybody?

Various pictures from Arny's posts

Reply #73
I looked at the Wayback Machine's files for PCABX.COM listening tests related to nonlinear distortion (the nonlinear piano) and found it to be missing some files and that other files were defective.

Here is hopefully a complete, correct set in FLAC format for your convenience:

[attachment=8304:piano_AM...s_FLACs_.zip]

The usual advice - start with the high order, high percentage distortion and work down.

Bold prediction: Everybody will hear one or more differences and in the end, some will be frustrated and others enlightened. ;-)

Various pictures from Arny's posts

Reply #74
Yes, I could recreate them, but I might screw that up in some subtle way. So if you would be so kind as to wait a week... Pretty Please?;-)


Yes, that's fine. No rush.

Quote
While I'm cleaning that project up, you said something about background noise. I think I can make that go away. Necessary to fix?


Do you mean by adding dither noise to both A and B as a masker to the existing residual background noises? That would make the most sense to me. If not too much trouble, please do. If you mean something else, please run it by me first, and hopefully I'll understand what you are suggesting.