I used audacity also.
At Analyse->Contrast there is a RMS level meter for the selection.
It gives -16.45 dB for track 7 and -16.80 dB for track 24.
I also did a sample difference between the tracks.
The spectrum of this difference has all audio frequencies, not only the shaped dither.
That might be expected. Running a file through a FIR will introduce a time delay in the output. If that delay is not accurately compensated for then you will see audio in the sample by sample difference.
The files look at least sample aligned.
A few samples delay offset would cause higher frequencies to end up in the difference signal. The lower freqencies would be still 'nulled' out if the levels are well matched.
I strongly suggest to listen and leave all analysing tools at home.
When playing the .wav files in JRiver and probably others too, you can let JRiver automatically adjust the volume setting for a track according to the EBU R128 algorithm when you absolutely want to exclude this.
However, this involves another digital processing step, so better leave this also for this test.
On top of that, the threshold for volume changes of our auditory system is supposed to be above 0.5dB, so 0.35dB should be no problem IMO.
Hans
When playing the .wav files in JRiver and probably others too, you can let JRiver automatically adjust the volume setting for a track according to the EBU R128 algorithm when you absolutely want to exclude this.
However, this involves another digital processing step, so better leave this also for this test.
On top of that, the threshold for volume changes of our auditory system is supposed to be above 0.5dB, so 0.35dB should be no problem IMO.
Hans
I'll look up the Lipshitz and Vanderkooy article where they claim that some people can hear 0.2 dB differences under double-blind conditions.
My ABX results for choosing the loudest track:
foo_abx 2.0.6c report
foobar2000 v1.4.4
2021-06-30 11:09:38
File A: 7 God Give Me Strength.wav
SHA1: a6b26bb33bd533adc2c8958a2afcf443580f738e
File B: 24 God Give Me Strength.wav
SHA1: 396d9aa117cc13150994d32832de502dd23b8179
Output:
DSD : WASAPI (event) : Speakers (2- SHD Series), 24-bit
Crossfading: NO
11:09:38 : Test started.
11:11:11 : 00/01
11:11:29 : 01/02
11:12:15 : 02/03
11:12:50 : 03/04
11:13:24 : 04/05
11:13:47 : 05/06
11:14:02 : 05/07
11:14:31 : 06/08
11:14:47 : 07/09
11:15:10 : 08/10
11:16:05 : 08/11
11:16:20 : 09/12
11:16:33 : 10/13
11:16:54 : 11/14
11:17:54 : 12/15
11:18:10 : 13/16
11:18:10 : Test finished.
----------
Total: 13/16
p-value: 0.0106 (1.06%)
-- signature --
48376dcc422387376386cc0c3f9265de8898ec8d
msv0,
With all respect, but I have no idea what you are trying to tell with the provided information, but it seems you want to prove something.
Is this in the interest of this listening test ?
If not, please concentrate on listening.
Hans
With all respect, but I have no idea what you are trying to tell with the provided information, but it seems you want to prove something.
Is this in the interest of this listening test ?
If not, please concentrate on listening.
Hans
Hi Hans,
It shows the level difference is audible for me and probably also for others.
This means the files are not usefull for checking if upsampling/downsampling is transparent as the level difference is too large.
It shows the level difference is audible for me and probably also for others.
This means the files are not usefull for checking if upsampling/downsampling is transparent as the level difference is too large.
Did you compare two files from Hans or just two files with 0.35 dB level difference? In the first case, how can you be sure it's the level difference that you hear?
When there is an audible volume difference, it does not automatically mean that the listening tests cannot be performed.
You have only checked one file, were figures as 0.35dB and now 1.06%(=0.51dB) were given.
Question is how are the other three files when tested the same way ?
When the loudness differences are known, the amp's volume level could be adjusted if needed.
So instead of problem finding, it would be more useful to report suggested volume adjustment solutions for the different files.
Hans
You have only checked one file, were figures as 0.35dB and now 1.06%(=0.51dB) were given.
Question is how are the other three files when tested the same way ?
When the loudness differences are known, the amp's volume level could be adjusted if needed.
So instead of problem finding, it would be more useful to report suggested volume adjustment solutions for the different files.
Hans
These were Hans's files.
I was listening for which one was louder.
The file that measured louder was also identified (in ABX) by me as louder.
But you are right: in theory there could be an other difference introduced by the proccessing that I perceive as a level difference... (But highly unlikely 🙂 )
I will zoom in today on the actual level difference, compensate for that and see if I can still hear a difference.
That 1.06 % is not the level difference, but the probability of having 13 out of 16 correct by mere chance.
How do you know that the difference you hear is due to the volume difference? Have you matched the 0.35dB level difference between, and then listened for whether you still hear a difference or not? Should you still hear a difference, then it's not due to the 0.35dB.
@ Hans I just PM'd you with my findings.
by the way, I converted the wav to apple lossless and in the TAG #ID the volumes of all tracks are the same
by the way, I converted the wav to apple lossless and in the TAG #ID the volumes of all tracks are the same
Ok will do the test.
I will add 0.35dB gain to file 24 and redo the ABX test.
In the mean time I did a more accurate test to find the level diffence, which confirms file 24 is 0.35dB lower in amplitude as file 7. (+/- 0.005 dB)
Just did did the test with +0.35dB gain on file 24:
foo_abx 2.0.6c report
foobar2000 v1.4.4
2021-06-30 14:01:14
File A: 7 God Give Me Strength.wav
SHA1: a6b26bb33bd533adc2c8958a2afcf443580f738e
File B: 24 God Give Me Strength 0_35dB.wav
SHA1: 8ed160bc69d75f4598f19fde48b88ef4910f34d6
Output:
DSD : WASAPI (event) : Speakers (2- SHD Series), 24-bit
Crossfading: NO
14:01:14 : Test started.
14:02:11 : 00/01
14:02:33 : 01/02
14:02:58 : 01/03
14:03:18 : 02/04
14:03:33 : 02/05
14:03:53 : 02/06
14:04:13 : 02/07
14:04:40 : 02/08
14:05:15 : 02/09
14:05:39 : 03/10
14:05:57 : 03/11
14:06:37 : 04/12
14:07:08 : 04/13
14:07:29 : 04/14
14:07:35 : 04/15
14:07:48 : 04/16
14:07:48 : Test finished.
----------
Total: 4/16
p-value: 0.9894 (98.94%)
-- signature --
9b7e032019b44647e174748f5dec9b4d05a0610c
Ok, so now we have a 98% chance that I dont know the difference.
So, looks like we confirmed with reasonable probability that:
1:There was a 0.35dB difference in the files 7 and 24
2: I could hear that difference
foo_abx 2.0.6c report
foobar2000 v1.4.4
2021-06-30 14:01:14
File A: 7 God Give Me Strength.wav
SHA1: a6b26bb33bd533adc2c8958a2afcf443580f738e
File B: 24 God Give Me Strength 0_35dB.wav
SHA1: 8ed160bc69d75f4598f19fde48b88ef4910f34d6
Output:
DSD : WASAPI (event) : Speakers (2- SHD Series), 24-bit
Crossfading: NO
14:01:14 : Test started.
14:02:11 : 00/01
14:02:33 : 01/02
14:02:58 : 01/03
14:03:18 : 02/04
14:03:33 : 02/05
14:03:53 : 02/06
14:04:13 : 02/07
14:04:40 : 02/08
14:05:15 : 02/09
14:05:39 : 03/10
14:05:57 : 03/11
14:06:37 : 04/12
14:07:08 : 04/13
14:07:29 : 04/14
14:07:35 : 04/15
14:07:48 : 04/16
14:07:48 : Test finished.
----------
Total: 4/16
p-value: 0.9894 (98.94%)
-- signature --
9b7e032019b44647e174748f5dec9b4d05a0610c
Ok, so now we have a 98% chance that I dont know the difference.
So, looks like we confirmed with reasonable probability that:
1:There was a 0.35dB difference in the files 7 and 24
2: I could hear that difference
I'm not sure what to make of this. The result is still far from the expected value for pure chance, which would be 8 out of 16 correct rather than only 4.