Skip to main content
Topic: Various pictures from Arny's posts (Read 36976 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Various pictures from Arny's posts

Reply #75
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.



Umm those background noises are dither, unnecessarily raised by about 20 dB.

Various pictures from Arny's posts

Reply #76
Here is hopefully a complete, correct set in FLAC format for your convenience:  [attachment=8304:piano_AM...s_FLACs_.zip]


Your tenfold jumps in distortion level are a little drastic. Listening to only the odd harmonic ones, 3 and 5, I found 10% distortion to be dead obvious [ABX logs will be available upon request, should this be TOS 8 challenged] but 1% was impossible. Do you have intermediate 3% distortion files handy? Don't knock yourself out to make them if you don't.

Quote
Umm those background noises are dither, unnecessarily raised by about 20 dB.


Then what were you proposing to do to erase the difference in background noise I sometimes hear in your various tests? What I'm suggesting is to put the reference file through the exact same processing chain that you do for the secondary files, ie never let us listeners even get access to the original master file at all, only a processed version of it.


Various pictures from Arny's posts

Reply #77
Here is hopefully a complete, correct set in FLAC format for your convenience:  [attachment=8304:piano_AM...s_FLACs_.zip]


Your tenfold jumps in distortion level are a little drastic. Listening to only the odd harmonic ones, 3 and 5, I found 10% distortion to be dead obvious [ABX logs will be available upon request, should this be TOS 8 challenged] but 1% was impossible. Do you have intermediate 3% distortion files handy? Don't knock yourself out to make them if you don't.

That can be done:  [attachment=8307:piano_AM_21_FLACs.zip]

[attachment=8306:piano_AM_21_FLACs.zip]

Quote
Umm those background noises are dither, unnecessarily raised by about 20 dB.


Then what were you proposing to do to erase the difference in background noise I sometimes hear in your various tests?


Just for the interchannel delay tests.

Quote
What I'm suggesting is to put the reference file through the exact same processing chain that you do for the secondary files, ie never let us listeners even get access to the original master file at all, only a processed version of it.


That is generally true of all of my samples.

Various pictures from Arny's posts

Reply #78
I might do better with sustained flute notes, since they don't have such a congested rich harmonic structure like piano does, however I can't hear 3% either. Oh well. This seems slightly strange to me since 10% is so blatantly obvious.


 


Various pictures from Arny's posts

Reply #79
I might do better with sustained flute notes, since they don't have such a congested rich harmonic structure like piano does, however I can't hear 3% either. Oh well. This seems slightly strange to me since 10% is so blatantly obvious.


Flutes tend to favor odd harmonics, and french horns seem to favor even harmonics.  Am I going to have to produce two more sets of files?

Various pictures from Arny's posts

Reply #80
Am I going to have to produce two more sets of files?


No need.

Out of curiosity, is the 5 second music excerpt from "In a restaurant by the sea" written by John Bucchino? Is this from some audio testing source analogous to the EBU SQAM files or did you just happen to pick it yourself?


Various pictures from Arny's posts

Reply #81
Am I going to have to produce two more sets of files?


No need.

Out of curiosity, is the 5 second music excerpt from "In a restaurant by the sea" written by John Bucchino? Is this from some audio testing source analogous to the EBU SQAM files or did you just happen to pick it yourself?


The original piano sample came from an unidentified segment of music that was a part of a 24/96 demo by a company that made ADCs and DACs for professionals at the time.

 

Various pictures from Arny's posts

Reply #82
Results of cubing test tones:

1 KHz test tone

[attachment=8308:Raw_cubed_1_KHz.png]

20 & 21 KHz twin tone

[attachment=8309:Raw_cube...0_21_kHz.png]

Various pictures from Arny's posts

Reply #83
Screen shots during development of nonlinear distortion test files:

Contents of test file during development (not finished yet)

[attachment=8310:test_gen..._annoted.png]

Results of processing on 1 KHz sine wave

[attachment=8313:test_gen...KHz_test.png]

1 kHz = -0.441 dB  2 kHz -44.8 dB actual THD 0.12 %

Results of processing 20 and 21 KHz twin tone

[attachment=8311:test_gen...KHz_test.png]

Checking on possible aliasing due to nonlinear processing in digital domain. Also for checking eventual downsampling.

Results of processing frequency response test (swish test)

[attachment=8312:test_gen...ish_test.png]

0.3 dB down at 20 Hz deemed to be inconsequential as critical 50Hz-15 KHz range is easily within 0.1 dB. This LF roll off due to limited FFT size processing in CEP interacting with use of 192 KHz sampling for actual processing.






Various pictures from Arny's posts

Reply #84
Origional file:

[attachment=8314:test_generation.png]

Even order file:

[attachment=8315:Raw_squared_.png]

Odd order file:

[attachment=8316:Raw_cubed_.png]

Admittedly these are dissimilar files, but doesn't the cubed file look much more like the original file than the squared file?

If you look at the 1 KHz tone as processed in the squared file, the fundamental has gone missing:

[attachment=8317:raw_squared_1_KHz.png]

If you look at the 20&21 kHz tones in the squared file, they have gone missing as well:

[attachment=8318:raw_squa...0_21_kHz.png]

Not so with the cubed file. All the original test tones are represented, albeit at somewhat changed but still highly significant levels.

For example, the 1 KHz segment of the cubed file:

[attachment=8319:Raw_cubed_1_KHz.png]


Various pictures from Arny's posts

Reply #85
Transformer-Based Speaker Simulators

[attachment=8322:speaker_simulators.png]


Various pictures from Arny's posts

Reply #87
Conference says:

"Upload failed. The file was larger than the available space"

This is a lie. The file is 6.1 megs. The available space is supposed to be 11.8 megs.  New math? ;-)

Various pictures from Arny's posts

Reply #88
My apologies. I went back to the same problem after a few reboots of windows 8.1 and now the troublesome zip file was 12.8 megs. The conference was right!

So I shortened the files.

These are the 24/96  ISC (Inter-Sample Clipping) test files that allow one to hear how files that contain no clipped samples can still sound clipped.

[attachment=8330:keys_jan..._shorter.zip]

Usual caveats. This is basically the keys jangling file with lots of ultrasonic content, so be careful when turning the volume up.  Delicate tweeters could be harmed, especially if your hearing is a little deficient at the high frequency end.  Headphones are recommended.

There are 3-4 instances of ISC clipping involving a very small number of samples distributed at approximately 2-3 second intervals throughout the file.

The file contains 9.5 seconds of the classic 24/96 keys jangling sound with and without intersample clipping. They are level matched, etc. and need no further processing.

Following the keys jangling sound is a low amplitude 1 second 4 KHz test tone to show that the primary listening test is over and the secondary listening test of the linearity of your monitoring system has begun.

You can use the FB2K file controls to keep from running into this part of the test until you have completed the primary test.

As you go back and forth between the two files from this point through the end of the file you should hear either nothing or a very low level rushing sound. Any other sound or audible difference between the two files indicates that your monitoring system is not of sufficient quality to be unconditionally undistorted when operated at 24/96.

I did do a FB2K listening test with 16/16 results but FB2K crashed while I was saving the log file.

Various pictures from Arny's posts

Reply #89
It is not intersample clipping one hears in the test file but soundwave discontinuity making a nasty pop sound. Same sound is present even when amplitude is lowered by several dBs more than the intersample peaks above full scale.

Various pictures from Arny's posts

Reply #90
It is not intersample clipping one hears in the test file but soundwave discontinuity making a nasty pop sound. Same sound is present even when amplitude is lowered by several dBs more than the intersample peaks above full scale.


Please describe the means you used to "lower the amplitude".

Various pictures from Arny's posts

Reply #91
Initially in foo_abx using the volume control in foobar2000. That works in digital domain and would remove clipping if it were present. Then opened the file in Adobe Audition CS6 and used Effects -> Amplitude and Compression -> Amplify... -> and set Gain to -15 dB for both channels.

Various pictures from Arny's posts

Reply #92
Initially in foo_abx using the volume control in foobar2000. That works in digital domain and would remove clipping if it were present.


That would be true if the function in question was immune to the effects of digital clipping. How do we know that this is true?


Quote
Then opened the file in Adobe Audition CS6 and used Effects -> Amplitude and Compression -> Amplify... -> and set Gain to -15 dB for both channels.


Now that is convincing  evidence!

I just did something similar and obtained similar results.  I have therefore confirmed your results by means of a far simpler path.  Just attentuate the file and play it by the simplest possible means.

Thank you for pointing out this fatal error in my first attempt at demonstrating this problem.

I was searching about for some means to validate the test and I thank you for your insights.

Back to the "Drawing Board".



Various pictures from Arny's posts

Reply #93
This is a updated/corrected version. The 27 step delay file is designed to be very easy and give you an idea of what to listen to.

[attachment=8333:impulses...rm_4416_.zip]

Test for 5, 7, 9 and 15 step delays were added to put in some more tests between doable and impossible.

The noise floor has been dropped dramatically by simply changing up the order of the last 2 operations.

The 0 sample delay file is your reference file  - compare all other files to it.

1 sample delay = 5.2 uSec
2 sample delay = 10.4 uSec
3 sample delay = 15.6 uSec
4 sample delay = 20.8 uSec
5 sample delay = 26.0 uSec
7 sample delay = 36.4 uSec
9 sample delay = 46.85 uSec - This is roughly the result that David L. Clark reported.
15 sample delay = 78.0 uSec
27 sample delay = 140.6 uSec - this is a confidence builder - you should be able to complete it easily and accurately. I do 16/16.  It is also a test of the suitability of your test environment.

Please do not proceed to the shorter delays until you can do well with the 27 step file.

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

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 normalized to 90% (-1 dB FS)
(6) All files were down sampled to 4416

Changing the order of 5 & 6 dramatically dropped the noise floor.

Various pictures from Arny's posts

Reply #94
Haven't tried 3 sample delay yet but 4 is doable with a nice quiet room. Unfortunaely there is traffic, trucks, and refrigerator noise here I have to contend with.

Code: [Select]

foo_abx 2.0 report
foobar2000 v1.3.3
2015-05-27 13:08:01

File A: Impulses shift 4 samples 2klp norm 4416 .flac
SHA1: 6133aaa124c97a3f768f3d9216af2eb07b7c0bf3
File B: Impulses shift 0 samples 2klp norm 4416 .flac
SHA1: 8fc00a4bb6a1bb0a66ec5c83cfaa36f9d8fddd13

Output:
DS : Primary Sound Driver
Crossfading: NO

13:08:01 : Test started.
13:08:56 : 00/01
13:09:11 : 01/02
13:09:33 : 02/03
13:09:46 : 03/04
13:10:11 : 04/05
13:10:32 : 05/06
13:11:10 : 06/07
13:11:20 : 07/08
13:11:47 : 08/09
13:12:07 : 09/10
13:12:07 : Test finished.

 ----------
Total: 9/10
Probability that you were guessing: 1.1%

 -- signature --
b5a468b183d7a98c037ae06ef08563400d89231b


Arny, are you sure David Clark's tests were with headphones, not speakers?
   


Various pictures from Arny's posts

Reply #95
3 takes more effort. My first stab at it, without training, was pretty much random but my second test was not as bad:

Code: [Select]

foo_abx 2.0 report
foobar2000 v1.3.3
2015-05-27 14:03:20

File A: Impulses shift 3 samples 2klp norm 4416 .flac
SHA1: 6fee6af7eedd131b9809333f86e5194539ca2f63
File B: Impulses shift 0 samples 2klp norm 4416 .flac
SHA1: 8fc00a4bb6a1bb0a66ec5c83cfaa36f9d8fddd13

Output:
DS : Primary Sound Driver
Crossfading: NO

14:03:20 : Test started.
14:04:49 : 01/01
14:05:49 : 02/02
14:06:13 : 02/03
14:06:28 : 03/04
14:06:46 : 04/05
14:07:49 : 05/06
14:08:17 : 06/07
14:08:34 : 07/08
14:09:22 : 07/09
14:10:25 : 08/10
14:10:25 : Test finished.

 ----------
Total: 8/10
Probability that you were guessing: 5.5%

 -- signature --
9f78862d1c5ea5ba107f26fe2337a82ae1778651

 


Various pictures from Arny's posts

Reply #96
2 is very hard.

Code: [Select]

foo_abx 2.0 report
foobar2000 v1.3.3
2015-05-27 14:19:43

File A: Impulses shift 2 samples 2klp norm 4416 .flac
SHA1: 0d55d7f19f67bcf33b9ca9416e9e7981e098ed03
File B: Impulses shift 0 samples 2klp norm 4416 .flac
SHA1: 8fc00a4bb6a1bb0a66ec5c83cfaa36f9d8fddd13

Output:
DS : Primary Sound Driver
Crossfading: NO

14:19:43 : Test started.
14:22:05 : 00/01
14:22:28 : 01/02
14:22:46 : 02/03
14:23:18 : 03/04
14:24:29 : 04/05
14:25:44 : 05/06
14:26:35 : 06/07
14:26:58 : 07/08
14:27:37 : 08/09
14:28:14 : 08/10
14:28:14 : Test finished.

 ----------
Total: 8/10
Probability that you were guessing: 5.5%

 -- signature --
852b926dbe8de4708857e9b624e2f167e8d621fa

My perception is that I switch from lateral soundstage localization cues to instead using tonal balance differences. One sound is more of a "thump" and the other is more of a "thunk" sound.

My understanding is that when we reach the JND points it is not at all uncommon for the listener to lose sight of what exactly it is that they are keying on. Is it level? Tone? Direction? Other?

 edit to add: Just tried 1 sample shift for yucks. No go.


Various pictures from Arny's posts

Reply #97
Haven't tried 3 sample delay yet but 4 is doable with a nice quiet room. Unfortunaely there is traffic, trucks, and refrigerator noise here I have to contend with.

Code: [Select]

foo_abx 2.0 report
foobar2000 v1.3.3
2015-05-27 13:08:01

File A: Impulses shift 4 samples 2klp norm 4416 .flac
SHA1: 6133aaa124c97a3f768f3d9216af2eb07b7c0bf3
File B: Impulses shift 0 samples 2klp norm 4416 .flac
SHA1: 8fc00a4bb6a1bb0a66ec5c83cfaa36f9d8fddd13

Output:
DS : Primary Sound Driver
Crossfading: NO

13:08:01 : Test started.
13:08:56 : 00/01
13:09:11 : 01/02
13:09:33 : 02/03
13:09:46 : 03/04
13:10:11 : 04/05
13:10:32 : 05/06
13:11:10 : 06/07
13:11:20 : 07/08
13:11:47 : 08/09
13:12:07 : 09/10
13:12:07 : Test finished.

 ----------
Total: 9/10
Probability that you were guessing: 1.1%

 -- signature --
b5a468b183d7a98c037ae06ef08563400d89231b


Arny, are you sure David Clark's tests were with headphones, not speakers?

Sorry if I conveyed the impression that Clark's ancient tests were done with headphones. Not true. It's been a while but my recollection is that the test rig was two midrange speakers on tracks that slid closer and farther from the listener, and the distance difference produced a delay that was part of the test. I think there may have been more to it than that, but it was definitely 2 speakers operated like that.

Looks like the cleaned up samples are giving you better results, no?

Various pictures from Arny's posts

Reply #98
Sensitivity using speakers instead of headphones should be lower so that would explain why my old ears are outdoing his test subjects.

Each time I take your tests I'm also getting training, so my better results may be due to that, I'm not sure. I wish other people would post their results so we could compare.

Clark's test for ITD by moving speakers  forwards and backwards is quite poor if you ask me, even if he compenstated for L vs. R level changes. Besides delay, using the method you described, he would have gotten differences in L vs R level, the acoustical power response measured at the ear would be different [nearfield response is dictated more heavily by direct, on-axis response and farfield sound is more heavily influenced by the summed reflected sound], plus there will be changes to the constructive/destructive comb filtering when any two speakers are separated by differing distances.





Various pictures from Arny's posts

Reply #99
Sensitivity using speakers instead of headphones should be lower so that would explain why my old ears are outdoing his test subjects.


My standard for comparison is my recollection of the results that you obtained with the last (noisy) set of samples. There was an outstanding request for a test involving what, 6 samples as that was around the place where you seemed to stop hearing differences. That's my recollection.

Now you are mucking around 2 samples. I'm coming up for my plan to do test files for < 1 sample. ;-)

Quote
Each time I take your tests I'm also getting training, so my better results may be due to that, I'm not sure.


Listener training works? That's not news! ;-)

Quote
I wish other people would post their results so we could compare.


I'd love to have other people try because the results of one listener don't have much weight.

I'd also like to have the samples themselves checked out for false positives, other errors.

Hopefully there will be people who are interested.  I think that very non-golden means have been used to produce fairly golden results.

 
SimplePortal 1.0.0 RC1 © 2008-2019