• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Is 22kHz brickwall cut off audible? - Listening test

Can you hear a difference between the files

  • I can hear a difference and I have an ABX result

    Votes: 2 11.8%
  • I can hear a difference but have no ABX result

    Votes: 3 17.6%
  • I cannot hear a difference and I have an ABX result

    Votes: 3 17.6%
  • I cannot hear a difference but have no ABX result

    Votes: 9 52.9%

  • Total voters
    17

krabapple

Major Contributor
Joined
Apr 15, 2016
Messages
1,922
Likes
2,031
10 trials is not enough but 2x10 trials is. It is an equivalent to 20 trials within some time interval. You may sum it and get 8/10 + 8/10 = 16/20. This is perfectly valid in probability theory and one may find what 16/20 means. BTW, I have a further 8/10 result. Why not 16 or 20 trials in one row? Because it is difficult to keep concentration, the differences are really very small. The only important outcome to me is, that 20kHz cut

...'done the way I did it' ...


may lead to a tiny sound difference and such difference is not distinguishable to everyone. Our hearing differs, and this is the fact. Humans are not measuring instruments, try to digest it. I have already posted the link that summarizes studies on about 400 subjects that confirms that some subjects are sensitive to >20kHz components that are accompanying the audio sound signals. There are no trivial answers to trivial questions, though some would like to have them. This is a bit funny, to me. Get humble, if you can.

you seem to see and hear what you want to -- whether it's in your experiments, or in what you read. This is becoming clear. The Reiss meta-analysis is not definitive.

If you really want to be truly scientific, now is the time to start figuring out *what* is causing the audible artifact you're reporting.
 

pjug

Major Contributor
Forum Donor
Joined
Feb 2, 2019
Messages
1,587
Likes
1,348
some subjects are sensitive to >20kHz components that are accompanying the audio sound signals.
I don't think the file created with your filter tests for this. You can hear music in the delta file when comparing 1 and 3. Whereas if you apply the filter like @pkane did in post #19, you don't hear anything when you play the delta. Can you filter your file such that you can't hear music in the delta and then try an ABX?
 

danadam

Addicted to Fun and Learning
Joined
Jan 20, 2017
Messages
655
Likes
956
10 trials is not enough but 2x10 trials is. It is an equivalent to 20 trials within some time interval.
True, but then we have to trust you, that you didn't do any trials in between.

I didn't listen even once :) :
foo_abx 2.0.6d report
foobar2000 v1.4.3
2022-07-12 22:46:33

File A: BreakOn_1sh.wav
SHA1: 73fd67f8fb135adf5e0ec92bc9f2178bbe269709
File B: BreakOn_3sh.wav
SHA1: 80aebddcdc26e9851a9441ddc89aae59a2d099eb

Output:
DS : Primary Sound Driver
Crossfading: NO

22:46:33 : Test started.
22:46:40 : 01/01
22:46:47 : 02/02
22:46:52 : 02/03
22:46:56 : 02/04
22:47:02 : 03/05
22:47:06 : 04/06
22:47:11 : 05/07
22:47:15 : 06/08
22:47:20 : 07/09
22:47:25 : 08/10
22:47:25 : Test finished.

----------
Total: 8/10
p-value: 0.0547 (5.47%)

-- signature --
6d677dca141bdd09fe6677d17c8f3d1541885dad
foo_abx 2.0.6d report
foobar2000 v1.4.3
2022-07-12 22:51:00

File A: BreakOn_1sh.wav
SHA1: 73fd67f8fb135adf5e0ec92bc9f2178bbe269709
File B: BreakOn_3sh.wav
SHA1: 80aebddcdc26e9851a9441ddc89aae59a2d099eb

Output:
DS : Primary Sound Driver
Crossfading: NO

22:51:00 : Test started.
22:51:08 : 01/01
22:51:14 : 01/02
22:51:20 : 01/03
22:51:26 : 02/04
22:51:33 : 03/05
22:51:39 : 04/06
22:51:45 : 05/07
22:51:51 : 06/08
22:51:57 : 07/09
22:52:03 : 08/10
22:52:03 : Test finished.

----------
Total: 8/10
p-value: 0.0547 (5.47%)

-- signature --
f8f0f02fe822fe2fc30631a1a4708cd3a3d47f00
foo_abx 2.0.6d report
foobar2000 v1.4.3
2022-07-12 23:00:29

File A: BreakOn_1sh.wav
SHA1: 73fd67f8fb135adf5e0ec92bc9f2178bbe269709
File B: BreakOn_3sh.wav
SHA1: 80aebddcdc26e9851a9441ddc89aae59a2d099eb

Output:
DS : Primary Sound Driver
Crossfading: NO

23:00:29 : Test started.
23:00:49 : 01/01
23:01:09 : 01/02
23:02:26 : 02/03
23:02:49 : 03/04
23:03:32 : 03/05
23:04:02 : 04/06
23:04:08 : 05/07
23:04:13 : 06/08
23:04:19 : 07/09
23:04:24 : 08/10
23:04:24 : Test finished.

----------
Total: 8/10
p-value: 0.0547 (5.47%)

-- signature --
17fd07ec9d7352796af606de2c0104d04e79dfb4
Why not 16 or 20 trials in one row? Because it is difficult to keep concentration, the differences are really very small.
You can always leave the test running until later.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,562
Likes
1,570
While it is pretty easy to create these positive results with pure luck or cheating (e.g. routing the output to a spectrum analyzer during tests), using this point as an argument will also discourage any potentially sincere ABX test in future.

I realized this long time ago and therefore rarely asked people to provide these online ABX results in case of such disputes.

I tried a different approach. Read the older and later replies to understand what the topic is about:

But unfortunately, this approach is not suitable for pma's topic because the steepness of filter transition will still be clearly visible in spectral analyses, even when multiple test files are provided.

One point I can agree is pma's positive results are not necessarily related to to ability to hear above 22kHz and that frequently posted AES paper.
 

rDin

Member
Joined
Jun 3, 2021
Messages
20
Likes
24
FWIW. The brickwalled file sounded "hard"/"clinical" to my ears, and so was easy enough to identify. Did 20 listens in 2 x 10 segments. Same results each time.

Capture.JPG
 

rDin

Member
Joined
Jun 3, 2021
Messages
20
Likes
24
Is that because BRICK WALL brings up images in your mind of hard objects and hard sounds? While soft gentle filters do the reverse.
The ABX software obfuscates the files so you don't know which is which. Before making my first choice, I listened to determine what differences I could hear, and then latched on to those differences during ABX testing. The most obvious difference was the "hard/clinical" nature of one file vs the other. After the test concluded and the results were revealed, those attributes I had latched on to were assigned to the brickwalled file. Could I have unconsciously assigned that particular labelling; sure, of course. But that didn't stop me from ABX'ing the differences - so whatever label I used, the differences were there...
 

AnalogSteph

Major Contributor
Joined
Nov 6, 2018
Messages
2,268
Likes
2,164
Location
.de
While listening on my JBL 104s (a DSP speaker) in the office next to active construction probably is a bit futile, I've tried another way of generating the lowpass-filtered file: Stacking two instances of the SoX resampler DSP in Foobar2000, the first resampling to 46000 Hz, the second back to 96000 Hz (otherwise default settings - Best, 95%, Linear Phase). Bit depth: auto, dither: never.

Let's see what DeltaWave 2.0.3 has to say.

Files provided by @pma:
delta-3sh.png

Note a slight response droop starting below 20 kHz.
pk-3sh.png

sd-3sh.png

dszoom-3sh.png

Note the tiniest bit of periodic filter ripple (maybe 0.00003 dB p-p).
Code:
DeltaWave v2.0.3, 2022-07-13T18:00:17.4968790+02:00
Reference:  BreakOn_1sh.wav[?] 5760000 samples 96000Hz 24bits, stereo, MD5=00
Comparison: BreakOn_3sh.wav[?] 5760000 samples 96000Hz 24bits, stereo, MD5=00
Settings:
    Gain:True, Remove DC:True
    Non-linear Gain EQ:False    Non-linear Phase EQ: False
    EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB
    Correct Non-linearity: False
    Correct Drift:True, Precision:30, Subsample Align:True
    Non-Linear drift Correction:False
    Upsample:False, Window:Hann
    Spectrum Window:Kaiser15, Spectrum Size:32768
    Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
    Filter Type:FIR, window:Hann, taps:8192, minimum phase=False
    Dither:False bits=0
    Trim Silence:False
    Enable Simple Waveform Measurement: False

Discarding Reference:  Start=0s, End=0s
Discarding Comparison: Start=0s, End=0s

Initial peak values Reference: -5,115dB   Comparison: -5,114dB
Initial RMS values Reference: -24,002dB   Comparison: -24,002dB

Null Depth=45,513dB
X-Correlation offset: 1 samples
Drift computation quality, #1: Excellent (0μs)


Trimmed 0 samples ( 0,00ms) front, 0 samples ( 0,00ms end)


Final peak values Reference: -5,115dB   Comparison: -5,115dB
Final RMS values Reference: -24,002dB   Comparison: -24,002dB

Gain= -0,0001dB (1x) DC=0 Phase offset=0,006379ms (0,612 samples)
Difference (rms) = -67,88dB [-116,61dBA]
Correlated Null Depth=90,42dB [113,26dBA]
Clock drift: 0 ppm


Files are NOT a bit-perfect match (match=8,28%) at 16 bits
Files are NOT a bit-perfect match (match=0,05%) at 24 bits
Files match @ 49,9944% when reduced to 12,36 bits


---- Phase difference (full bandwidth): 82,5165000366631°
    0-10kHz: 0,00°
    0-20kHz: 0,01°
    0-24kHz: 30,40°
Timing error (rms jitter): 65,8ns
PK Metric (step=400ms, overlap=50%):
RMS=-114,3dBr
Median=-118,6
Max=-89,0

99%: -111,51
75%: -115,97
50%: -118,58
25%: -122,32
1%: -132,87

gn=1,00000754484096, dc=-5,87161240021227E-08, dr=0, of=0,6124191116

DONE!

Signature: 4073a37c71102eb998bb8203847305fc

RMS of the difference of spectra: -136,357137763593dB
DF Metric (step=400ms, overlap=0%):
Median=-50,9dB
Max=-25,9dB Min=-57,7dB

1% > -57,52dB
10% > -54,51dB
25% > -53,44dB
50% > -50,88dB
75% > -47,02dB
90% > -43,26dB
99% > -21,13dB

Linearity 24,5bits @ 0.5dB error

As opposed to mine:
delta-46.png

I'll admit, it's a bit less steep.
pk-46.png

sd-46.png

1657728115090.png

Zero discernable passband ripple in sight here.

Code:
DeltaWave v2.0.3, 2022-07-13T17:40:25.6409122+02:00
Reference:  BreakOn_1sh.wav[?] 5760000 samples 96000Hz 24bits, stereo, MD5=00
Comparison: BreakOn_1sh-46k96.flac[?] 5760000 samples 96000Hz 24bits, stereo, MD5=00
Settings:
    Gain:True, Remove DC:True
    Non-linear Gain EQ:False    Non-linear Phase EQ: False
    EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB
    Correct Non-linearity: False
    Correct Drift:True, Precision:30, Subsample Align:True
    Non-Linear drift Correction:False
    Upsample:False, Window:Hann
    Spectrum Window:Kaiser15, Spectrum Size:32768
    Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
    Filter Type:FIR, window:Hann, taps:8192, minimum phase=False
    Dither:False bits=0
    Trim Silence:False
    Enable Simple Waveform Measurement: False

Discarding Reference:  Start=0s, End=0s
Discarding Comparison: Start=0s, End=0s

Initial peak values Reference: -5,115dB   Comparison: -5,118dB
Initial RMS values Reference: -24,002dB   Comparison: -24,002dB

Null Depth=66,091dB
X-Correlation offset: 0 samples
Drift computation quality, #1: Excellent (0μs)


Trimmed 0 samples ( 0,00ms) front, 0 samples ( 0,00ms end)


Final peak values Reference: -5,115dB   Comparison: -5,118dB
Final RMS values Reference: -24,002dB   Comparison: -24,002dB

Gain= 0dB (1x) DC=0 Phase offset=0ms (0 samples)
Difference (rms) = -69,95dB [-153,33dBA]
Correlated Null Depth=116,2dB [144,63dBA]
Clock drift: 0 ppm


Files are NOT a bit-perfect match (match=9,29%) at 16 bits
Files are NOT a bit-perfect match (match=0,1%) at 24 bits
Files match @ 49,9767% when reduced to 12,63 bits


---- Phase difference (full bandwidth): 68,1650883982791°
    0-10kHz: 0,00°
    0-20kHz: 0,00°
    0-24kHz: 24,20°
Timing error (rms jitter): 51,1ns
PK Metric (step=400ms, overlap=50%):
RMS=-138,1dBr
Median=-142,5
Max=-123,8

99%: -124,68
75%: -141,73
50%: -142,46
25%: -143,33
1%: -146,23

gn=1,00000000100437, dc=-5,58489578737606E-13, dr=0, of=0

DONE!

Signature: cbdb97725bcf9a056a8160301da46fac

RMS of the difference of spectra: -153,022741548246dB
DF Metric (step=400ms, overlap=0%):
Median=-52,3dB
Max=-26dB Min=-59dB

1% > -58,34dB
10% > -55,94dB
25% > -54,55dB
50% > -52,31dB
75% > -48,67dB
90% > -44,95dB
99% > 0,0dB

Linearity 24,6bits @ 0.5dB error
Note: The DeltaWave changelog notes improved sub-sample offset accuracy for v2.0.0, so an update is advised. I'm impressed by what you can coax out of it these days.

BTW, I was trying an intermediate 46102 Hz at first, that resulted in linearly ascending clock drift of a substantial degree and generally worse results. Perhaps that was asking a bit much even from a resampling algorithm supporting "arbitrary" ratios on paper, or is challenging even the improved sub-sample offset correction.
 

Attachments

  • dszoom-46.png
    dszoom-46.png
    66.5 KB · Views: 10
Last edited:

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,562
Likes
1,570
BTW, I was trying an intermediate 46102 Hz at first, that resulted in linearly ascending clock drift of a substantial degree and generally worse results. Perhaps that was asking a bit much even from a resampling algorithm supporting "arbitrary" ratios on paper, or is challenging even the improved sub-sample offset correction.
Probably uninteresting to other thread participants, foobar's old PPHS resampler can get stupidly slow when resample to and from prime numbers, also not so good artifact suppression. SSRC on the other hand does not support prime numbers at all.

RetroArch can deal with prime numbers with good quality and reasonable speed. It probably has some SIMD optimization as well, but still slower than the SoX resampler plugins optimized by lvqcl.

My way to evaluate quality is a simple approach: two instances of resampler plugin to round trip the standard RMAA signal set (e.g. 44100 -> 76543 -> 44100), and the test suite from Infinite Wave website. So no null test, just spectral analyses.
 

levimax

Major Contributor
Joined
Dec 28, 2018
Messages
1,502
Likes
2,137
Location
San Diego
Probably uninteresting to other thread participants, foobar's old PPHS resampler can get stupidly slow when resample to and from prime numbers, also not so good artifact suppression. SSRC on the other hand does not support prime numbers at all.

RetroArch can deal with prime numbers with good quality and reasonable speed. It probably has some SIMD optimization as well, but still slower than the SoX resampler plugins optimized by lvqcl.

My way to evaluate quality is a simple approach: two instances of resampler plugin to round trip the standard RMAA signal set (e.g. 44100 -> 76543 -> 44100), and the test suite from Infinite Wave website. So no null test, just spectral analyses.
I use the Resampler -V plugin and select SOX resampling.... do you recommend RetroArch instead?
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,562
Likes
1,570
I use the Resampler -V plugin and select SOX resampling.... do you recommend RetroArch instead?
I use this SoX plugin (foo_dsp_resampler)
So, not the SoX from Resampler-V. I have not tested Resampler-V therefore I have no comment.

In terms of speed and measured quality, I would say foo_dsp_resampler is better than RetroArch.
Speed is important for me because I often use the resampler plugin in ReplayGain's True Peak scan.
 
Top Bottom