Audibility of low frequency distortion in speakers

[flipflop]

Spoiler: TA2 ABX log

foo_abx 2.0.6c report
foobar2000 v1.3.16
2020-12-14 01:34:52

File A: TA2 clean.wav
SHA1: 886f9c86ae404b63d3d5a76f0e23914aff2432cb
File B: TA2 distorted.wav
SHA1: c44c63ea81ce0457717416815caaee38ec655cb5

Output:
DS : Primær lyddriver
Crossfading: NO

01:34:52 : Test started.
01:35:10 : 01/01
01:35:17 : 02/02
01:35:26 : 03/03
01:35:35 : 04/04
01:35:46 : 05/05
01:35:59 : 06/06
01:36:14 : 07/07
01:36:19 : 08/08
01:36:29 : 09/09
01:36:46 : 10/10
01:37:00 : 11/11
01:37:05 : 12/12
01:37:05 : Test finished.

----------
Total: 12/12
p-value: 0.0002 (0.02%)

-- signature --
d680deccdc6605c3cf12c80460b6d228c446c018

 

[March Audio]

Without all of us listening to the examples at roughly the same sound pressure level, it doesn't really make sense after all.

At 10dB sound pressure difference, the audibility of the distortion in percent can change by a factor of three.

At the very least, everyone should estimate at what sound pressure level the listening test was performed and include this in the result.

Would it make sense to provide a kind of calibration tone, recorded for example at -80dB or -70dB, and then adjust your speakers or headphones so that the sound is just perceived?
Because then we would have something like a common sensation level.

This calibration tone is then, with a short instruction, simply added to each example.

As already discussed, yes the replay volume does make a difference. However in the context of this test where we are not trying to examine what is the limit of audibility it is not a significant issue.

The limit of audibility does indeed vary with volume, but even so its not so much as to make it significant for the purposes of this test.

Highlighted below the difference in the limit of audibilty at 50Hz between 80dB and 100dB. Directly below it is for 100Hz which is a higher frequency than I will be testing.

However for those that have the capability to measure the SPL the yes its an excellent idea of yours to provide a calibration tone. People could use their phones and an SLM app to get a rough idea of the level.

In fact just play the "clean" file and set to the volume to 80 to 85dB.

Any recommendations for a good Android and Apple apps?

 

[mcdn]

Ok first real test (A2)

I will show the distortion levels and plot of the FFT later once some results have come in.

A tip. If the difference is obvious to you then you can speed up the test by just comparing X and Y. The order of A and B remains constant

No foobar plugins on Mac, so using squishyball, thanks for the link @q3cpma.

matt@vasse squishyball % ./squishyball --abx TestA2/TA2\ clean.wav TestA2/TA2\ distorted.wav

A/B/X test results:
    Correct sample identified 20/20 trials.
    Probability of 20 correct via random chance: 0.00%
    Statistically significant result (>=99% confidence).

Testing metadata:
    Total time spent testing: 2:32.90
    Total seeks: 1
    Silent flip used 52 times.
    Undo was not used.

The difference between tracks was blindingly obvious on Sennheiser HD558s. Will try loudspeakers later.

 

[March Audio]

Just to expand on my comments above as to the purpose of this part of tests (listening to tones). As mentioned its not to establish the limits of audibility. Its to see/confirm if people can reliably identify distortion a little above the limits shown in the AES article shown above. It also gives people who have not done so before, an opportunity to play with Foobar and get used to ABX testing.

The other thing it does is provide a little training as to what this harmonic distortion may sound like. Often with music the distortion isnt obvious as distortion. It is often a subtle change in timbre. A sharpening of edges (transients), an emphasis or brightening of certain aspect of a struck note.
At lower frequencies this can be a "thickening" of the sound.

 

[March Audio]

Regarding phone SLM meters the Android app I use iNVH by Bosch. I dont know if its available on ios. With my Samsung s20 out of the box its not million miles out but can be calibrated.


EDIT: just checked, it is available on ios

 

[mcdn]

Regarding phone SLM meters the Android app I use iNVH by Bosch. I dont know if its available on ios. With my Samsung s20 out of the box its not million miles out but can be calibrated.

Hmm, a quick and dirty test suggests more investigation needed before trusting any of these apps: Accuracy of phone based SPL meters

 

[March Audio]

Hmm, a quick and dirty test suggests more investigation needed before trusting any of these apps: Accuracy of phone based SPL meters

They need to be calibrated and will still have significant limitations due to the hardware. Its only for ball park. I will post some further data/tests on iNVM later.

 

[dougi]

Are people hearing the 2.8% distortion in the second video?

Nope, not on crappy work headset. The 8.5% was obvious.

 

[Blumlein 88]

Done on laptop with headphones because it was convenient.

Spoiler

foo_abx 2.0.6d report
foobar2000 v1.5.3
2020-12-13 22:19:29

File A: TA2 clean.wav
SHA1: 886f9c86ae404b63d3d5a76f0e23914aff2432cb
File B: TA2 distorted.wav
SHA1: c44c63ea81ce0457717416815caaee38ec655cb5

Output:
DS : Speakers (Realtek High Definition Audio)
Crossfading: NO

22:19:29 : Test started.
22:21:15 : 01/01
22:21:38 : 02/02
22:21:52 : 03/03
22:22:18 : 04/04
22:23:44 : 04/05
22:24:13 : 04/06
22:24:55 : 05/07
22:25:21 : 06/08
22:25:46 : 07/09
22:26:26 : 07/10
22:26:36 : 08/11
22:26:55 : 09/12
22:27:14 : 09/13
22:27:28 : 10/14
22:27:46 : 11/15
22:27:56 : 12/16
22:27:56 : Test finished.

----------
Total: 12/16
p-value: 0.0384 (3.84%)

-- signature --

 

[Doodski]

I opened the (A2) ABX test and compared the X & Y about 15 times and found absolutely no difference so I never completed the 16 tests.

 

[mcdn]

The difference between tracks was blindingly obvious on Sennheiser HD558s. Will try loudspeakers later.

TestA2 again, and this time on loudspeakers. 19/20, but really 20/20 because I just accidentally hit the wrong button.

matt@vasse squishyball % ./squishyball --abx TestA2/TA2\ clean.wav TestA2/TA2\ distorted.wav

A/B/X test results:
    Correct sample identified 19/20 trials.
    Probability of 19 or better correct via random chance: 0.00%
    Statistically significant result (>=99% confidence).

Testing metadata:
    Total time spent testing: 1:57.90
    Total seeks: 0
    Silent flip used 47 times.
    Undo was not used.

 

[mcdn]

I opened the (A2) ABX test and compared the X & Y about 15 times and found absolutely no difference so I never completed the 16 tests.

Have you checked you can tell the difference with the TestA1.zip files? If you can't hear that then there is something up with your setup.

 

[Doodski]

Have you checked you can tell the difference with the TestA1.zip files? If you can't hear that then there is something up with your setup.

Yes, I completed the A1 test with 16 of 16 correct. See post #93
https://www.audiosciencereview.com/...stortion-in-speakers.18291/page-5#post-599040

 

[mcdn]

Yes, I completed the A1 test with 16 of 16 correct. See post #93
https://www.audiosciencereview.com/...stortion-in-speakers.18291/page-5#post-599040

My bad, should have checked before asking. How interesting.

 

[March Audio]

Yes, I completed the A1 test with 16 of 16 correct. See post #93
https://www.audiosciencereview.com/...stortion-in-speakers.18291/page-5#post-599040

Try playing A and then press B whilst it is only say 1/2 way through so the switch is as seemless as possible. Then try it the other way B then A. The difference may then become easier to spot. Often its just knowing what to look for

 

[Doodski]

My bad, should have checked before asking. How interesting.

Yes, it is interesting. I disabled the EQ, no sound effects enabled (I never use them) and volume to the max and no difference.

 

[Doodski]

OK redid it according to your recommendation to instantly switch from X to Y and I did better.

 

[mcdn]

Try playing A and then press B whilst it is only say 1/2 way through so the switch is as seemless as possible. Then try it the other way B then A. The difference may then become easier to spot. Often its just knowing what to look for

I just did some switching between A&B on the first trial to start with, then went to X, decided what I think it is (A or B), switched to that to confirm and then moved to the next trial. If it's really easy like with the DA2 files I just hit X on each subsequent trial, quick switch to A or B depending, and then confirm.

Is that how foobar2000's ABX works as well?

 

[Doodski]

When I did the A1 test with 16 of 16 correct I noticed that A & B never changed position.

 

[Blumlein 88]

Okay, did the test over speakers, and it is much more obvious. I guess the question is whether the speakers were adding their own distortion more than the phones were. Or perhaps the phones were distorting enough it masked the cleaner of the signals.

Spoiler

foo_abx 2.0.6d report
foobar2000 v1.5.3
2020-12-14 00:57:30

File A: TA2 clean.wav
SHA1: 886f9c86ae404b63d3d5a76f0e23914aff2432cb
File B: TA2 distorted.wav
SHA1: c44c63ea81ce0457717416815caaee38ec655cb5

Output:
DS : marantz-AVR (Intel(R) Display Audio)
Crossfading: NO

00:57:30 : Test started.
00:59:42 : 01/01
01:00:07 : 02/02
01:00:34 : 03/03
01:00:59 : 03/04
01:01:22 : 04/05
01:01:40 : 05/06
01:01:57 : 06/07
01:02:28 : 07/08
01:03:11 : 08/09
01:03:25 : 09/10
01:03:36 : 10/11
01:03:46 : 11/12
01:03:56 : 12/13
01:04:36 : 13/14
01:05:11 : 14/15
01:05:33 : 15/16
01:05:33 : Test finished.

----------
Total: 15/16
p-value: 0.0003 (0.03%)

-- signature --
b66a2bd41ff6fc6381656874802c2da2ef255624