Skip to main content
Topic: ABX test of frequency discrimination (Read 4510 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

ABX test of frequency discrimination

It is well known that the average just-noticeable difference (JND) for a 500 Hz tone is about 0.5%, but as low as 1 Hz (0.2%) in <50 year old adults[1].

I want to see what success rates people achieve by using the foobar2000 ABX comparator.

Please also consider this:
- use 13 trials,
- listen to the files in their entirety, don't use the set start/stop feature
- post if you are younger or older than 50 years (you can be more specific if you want), because you will probably fail if you're older

Why 13 trials?
A p-value of <5% is reached with 10/13, which is the shortest test that coincides approximately with a 75% success rate.

What to listen for?
The 501 file contains one tone that has higher frequency (501 Hz instead of 500 Hz).


Download: test files

I'm awaiting the logs. (post them in [code or codebox] tags)
If you gave your best and didn't achieve <5% then please post your results anyway!




[1] Bungert-Kahl P., Biedermann F., Dörrscheidt G. J., von Cramon D. Y., Rübsamen R. (2004). Psychoacoustic test tools for the detection of deficits in central auditory processing: normative data. Z. Audiol. 43, 48–71
[2] Freigang C., Schmidt L., Wagner J., Eckardt R., Steinhagen-Thiessen E., Ernst A., Rübsamen R. (2011). Evaluation of central auditory discrimination abilities in older adults. Front Aging Neurosci. 3(6)
"I hear it when I see it."

ABX test of frequency discrimination

Reply #1
Here's my first attempt:

I'm <30 years old.
Code: [Select]
foo_abx 2.0.1 report
foobar2000 v1.3.8
2015-09-27 17:25:30

File A: jnd-freq-500.flac
SHA1: 90c79cc5d38b363d79cd02b219a6b569070ad7f0
File B: jnd-freq-501.flac
SHA1: 0cd6681fba027ded6cd4a7e3b507da3534323227

Output:
DS : Primärer Soundtreiber
Crossfading: NO

17:25:30 : Test started.
17:25:42 : 01/01
17:25:49 : 02/02
17:26:00 : 03/03
17:26:09 : 03/04
17:26:17 : 04/05
17:26:25 : 05/06
17:26:34 : 06/07
17:26:40 : 07/08
17:26:46 : 08/09
17:27:03 : 09/10
17:27:12 : 10/11
17:27:22 : 11/12
17:27:32 : 12/13
17:27:32 : Test finished.

 ----------
Total: 12/13
Probability that you were guessing: 0.2%

 -- signature --
089ee13961595fe3ddd5ecb400acd2a47a382d26
"I hear it when I see it."

ABX test of frequency discrimination

Reply #2
I forgot to set it to 13 trials, but here it is:
(I'm less than 50 years old.)

Code: [Select]
foo_abx 2.0 report
foobar2000 v1.3.8
2015-09-27 19:13:07

File A: jnd-freq-500.flac
SHA1: 90c79cc5d38b363d79cd02b219a6b569070ad7f0
File B: jnd-freq-501.flac
SHA1: 0cd6681fba027ded6cd4a7e3b507da3534323227

Output:
DS : Primary Sound Driver
Crossfading: NO

19:13:07 : Test started.
19:15:46 : 01/01
19:15:55 : 01/02
19:16:02 : 02/03
19:16:13 : 03/04
19:16:52 : 04/05
19:17:29 : 05/06
19:17:41 : 06/07
19:17:48 : 07/08
19:18:16 : 08/09
19:18:58 : 09/10
19:19:17 : 10/11
19:19:30 : 11/12
19:19:52 : 12/13
19:20:03 : 12/14
19:20:21 : 12/15
19:20:46 : 12/16
19:20:46 : Test finished.

  ----------
Total: 12/16
Probability that you were guessing: 3.8%

  -- signature --
f7c5006d2b41fc1c15c67840632e6d13afe1c072

ABX test of frequency discrimination

Reply #3
Thanks. Judging from the timestamps you didn't just skip the last 3 "extra" trials, but gave your best, so they should be counted?

But the result is significant anyway and 16 trials works too.
"I hear it when I see it."

ABX test of frequency discrimination

Reply #4
Don't know if you are doing this intentionally or not but a continuous tone is much easier than separated short tones like your files. I am younger than 50.

Code: [Select]
foo_abx 2.0.1 report
foobar2000 v1.3.8
2015-09-28 04:04:52

File A: jnd-freq-500.flac
SHA1: 90c79cc5d38b363d79cd02b219a6b569070ad7f0
File B: jnd-freq-501.flac
SHA1: 0cd6681fba027ded6cd4a7e3b507da3534323227

Output:
ASIO : Creative ASIO
Crossfading: NO

04:04:52 : Test started.
04:06:06 : 01/01
04:06:23 : 02/02
04:07:13 : 02/03
04:07:55 : 03/04
04:09:07 : 04/05
04:09:20 : 05/06
04:09:37 : 06/07
04:09:48 : 07/08
04:10:00 : 08/09
04:10:29 : 09/10
04:10:43 : 10/11
04:11:12 : 10/12
04:11:26 : 11/13
04:11:26 : Test finished.

----------
Total: 11/13
Probability that you were guessing: 1.1%

-- signature --
c9ff410d625ef2363c63263b4d5452fca21ad9c4

ABX test of frequency discrimination

Reply #5
Thanks. Judging from the timestamps you didn't just skip the last 3 "extra" trials, but gave your best, so they should be counted?

Yeah they should be counted. I might also add that I was never sure I got a comparison right, just a slight "feeling".

ABX test of frequency discrimination

Reply #6
@bennetng: Thanks! Yes it is intentional, it' not supposed to be easy.

@Brand: Alright, that's what I thought. Yeah, you are not supposed to hear a clear difference but a just-noticeable difference (JND).
"I hear it when I see it."

ABX test of frequency discrimination

Reply #7
It is well known that the just-noticeable difference (JND) for a 500 Hz tone is about 1 Hz.


Really?

I'm under the impression that a JND is 0.5%, which  at 500 Hz is more like 2.5 Hz.

www.britannica.com frequency-just-noticeable-difference

"The audio frequency range encompasses nearly nine octaves. Over most of this range, the minimum change in the frequency of a sinusoidal tone that can be detected by the ear, called the frequency just noticeable difference, is about 0.5 percent of the frequency of the tone, or about one-tenth of a musical half-step. The ear is less sensitive near the upper and lower ends of the audible spectrum"

washington.edu.

Gives the JND @ 520 Hz as about 2 Hz.

There are other references such as Zwicker and Fastl that disagree, but that leads me to conclude that this is not a well-known fact.  Some controversy seems to remain.

ABX test of frequency discrimination

Reply #8
It is well known that the just-noticeable difference (JND) for a 500 Hz tone is about 1 Hz.


Really?

I'm under the impression that a JND is 0.5%, which  at 500 Hz is more like 2.5 Hz.

www.britannica.com frequency-just-noticeable-difference

"The audio frequency range encompasses nearly nine octaves. Over most of this range, the minimum change in the frequency of a sinusoidal tone that can be detected by the ear, called the frequency just noticeable difference, is about 0.5 percent of the frequency of the tone, or about one-tenth of a musical half-step. The ear is less sensitive near the upper and lower ends of the audible spectrum"

washington.edu.

Gives the JND @ 520 Hz as about 2 Hz.

There are other references such as Zwicker and Fastl that disagree, but that leads me to conclude that this is not a well-known fact.  Some controversy seems to remain.


Wasn't there a web page with an app that could test this?  I remember having it well under 1Hz in the 400 - 500Hz range, which put me in the top <1% or something.  Of course, it took some practice to achieve that result.

ABX test of frequency discrimination

Reply #9
I'm under the impression that a JND is 0.5%, which  at 500 Hz is more like 2.5 Hz.

Even after three people just indicated they could discriminate a 1 Hz difference???
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

ABX test of frequency discrimination

Reply #10
It is well known that the just-noticeable difference (JND) for a 500 Hz tone is about 1 Hz.

I want to see what success rates people achieve by using the foobar2000 ABX comparator.

Please also consider this:
- use 13 or 16 trials,
- listen to the files in their entirety, don't use the set start/stop feature
- post if you are younger or older than 50 years (you can be more specific if you want), because you will probably fail if you're older

Why 13 trials?
A p-value of <5% is reached with 10/13, which is the shortest test that coincides approximately with a 75% success rate.

What to listen for?
The 501 file contains one tone that has higher frequency (501 Hz instead of 500 Hz).


Download: test files

I'm awaiting the logs. (post them in [code or codebox] tags)


Did you try the adaptive pitch test on this page?

tonometric.com/adaptivepitch


ABX test of frequency discrimination

Reply #12
@Arnold: You are right if we're talking about an average. I've edited the main post and added a citation. But I want to look at this "limit" and see what ABX scores people produce.


To anyone else, if you gave your best and couldn't achieve <5% then please go ahead and post the log anyway! Actually, I expect more such results.
"I hear it when I see it."

ABX test of frequency discrimination

Reply #13
Did you try the adaptive pitch test on this page?

tonometric.com/adaptivepitch

Confident up to 0.75, failed at 0.375

Just want to clarify I misunderstood the results. I even reached 0.1875 during the test but the final result said 1.125 Hz.


ABX test of frequency discrimination

Reply #15
Just want to clarify I misunderstood the results. I even reached 0.1875 during the test but the final result said 1.125 Hz.


That's the nature of such up-down or staircase tests. Even if you don't hear a difference anymore (but very well may think you do) there is a 50% chance of going down a step. So by chance alone you could go down a few steps, which is why there's a suggestion to have at least a few reversals before such a tests can end.

Obviously, you cannot just use those lowest steps as evidence for audibility.
"I hear it when I see it."

ABX test of frequency discrimination

Reply #16
@Arnold: You are right if we're talking about an average. I've edited the main post and added a citation. But I want to look at this "limit" and see what ABX scores people produce.


To anyone else, if you gave your best and couldn't achieve <5% then please go ahead and post the log anyway! Actually, I expect more such results.

tips about tricks or tips for obtaining better scores (that aren't cheating)?

ABX test of frequency discrimination

Reply #17
I'm not sure, I just tried again and didn't score better than in my first log. But I noticed that at "higher" SPL I score worse.

So maybe try a lower volume. Also try breaks, listening to these tones over and over again can be very fatiguing, more so at higher SPL.
Most importantly, keep in mind that each file has 3 short tones, but only one of them in the 501 file has actually higher frequency.

But, above a certain age, it may just be impossible comparable to trying to hear 19 kHz with HF hearing loss.
"I hear it when I see it."

ABX test of frequency discrimination

Reply #18
Just want to clarify I misunderstood the results. I even reached 0.1875 during the test but the final result said 1.125 Hz.


That's the nature of such up-down or staircase tests. Even if you don't hear a difference anymore (but very well may think you do) there is a 50% chance of going down a step. So by chance alone you could go down a few steps, which is why there's a suggestion to have at least a few reversals before such a tests can end.

Obviously, you cannot just use those lowest steps as evidence for audibility.


Is it possible to hear a difference but not be able to tell if it's higher or lower?

ABX test of frequency discrimination

Reply #19
Is it possible to hear a difference but not be able to tell if it's higher or lower?


That's an interesting question. Now I could just say that e.g. in the case of a higher frequency the difference is an increase in frequency and therefore if you hear a difference, it is this increase in frequency that you hear.

The problem in the adaptive pitch test is that you know there is a difference. 100%. There are no 2 equal tones trials. The second tone is always higher OR lower. You know that and so you probably think you can hear a difference, even if you have surpassed the limits of your perception.

So I would say that the inability to hear whether it is higher or lower in that adaptive test means that you do not hear a difference. Otherwise you would always be right.


I think there is another problem with this adaptive test: consecutive trials combined with auditory memory make it easier to hear a difference. Let's say you are at +1 Hz and are at the limit of your perception, but guess/hear right, then you can immediately listen to the next trial with either the same +1 Hz or actually -2 Hz relative to the second tone in the previous trial.
"I hear it when I see it."

ABX test of frequency discrimination

Reply #20
Quote
If you gave your best and didn't achieve <5% then please post your results anyway! wink.gif


Code: [Select]
foo_abx 2.0.1 report
foobar2000 v1.3.8
2015-09-29 00:07:09

File A: jnd-freq-500.flac
SHA1: 90c79cc5d38b363d79cd02b219a6b569070ad7f0
File B: jnd-freq-501.flac
SHA1: 0cd6681fba027ded6cd4a7e3b507da3534323227

Output:
DS : Driver de som primário
Crossfading: NO

00:07:09 : Test started.
00:08:02 : 01/01
00:08:26 : 01/02
00:08:35 : 01/03
00:08:42 : 01/04
00:08:50 : 01/05
00:08:57 : 02/06
00:09:04 : 03/07
00:09:12 : 03/08
00:09:19 : 04/09
00:09:25 : 04/10
00:09:32 : 05/11
00:09:41 : 05/12
00:09:47 : 06/13
00:09:47 : Test finished.

----------
Total: 6/13
Probability that you were guessing: 70.9%

-- signature --
dac44176409f5c171a6214de4ded84eb9199b0b4


I'm 25.

ABX test of frequency discrimination

Reply #21
Just want to clarify I misunderstood the results. I even reached 0.1875 during the test but the final result said 1.125 Hz.


That's the nature of such up-down or staircase tests. Even if you don't hear a difference anymore (but very well may think you do) there is a 50% chance of going down a step. So by chance alone you could go down a few steps, which is why there's a suggestion to have at least a few reversals before such a tests can end.

Obviously, you cannot just use those lowest steps as evidence for audibility.


Is it possible to hear a difference but not be able to tell if it's higher or lower?


That is IME an observable contrition with most if not all audible differences. Near threshold, I can't tell what the difference is, but I know it is a difference and score reliably.

BTW I last did JND tests related to frequency variations about 5 years ago. I could at that time confirm the 0.5% value. Not any more. I'm random guessing at 1%.  Old age sucks!

ABX test of frequency discrimination

Reply #22
Contrition? What are you talking about?
"I hear it when I see it."

ABX test of frequency discrimination

Reply #23
I didn't think I heard any differences at any point and thought I was wasting my time.
Code: [Select]
foo_abx 2.0.1 report
foobar2000 v1.3.9 beta 3
2015-09-29 21:45:30

File A: jnd-freq-500.flac
SHA1: 90c79cc5d38b363d79cd02b219a6b569070ad7f0
File B: jnd-freq-501.flac
SHA1: 0cd6681fba027ded6cd4a7e3b507da3534323227

Output:
WASAPI (event) : Speakers (XMOS USB Audio), 24-bit
Crossfading: NO

21:45:30 : Test started.
21:45:59 : 00/01
21:46:20 : 01/02
21:46:35 : 02/03
21:46:49 : 03/04
21:47:03 : 03/05
21:47:24 : 04/06
21:47:38 : 05/07
21:48:10 : 06/08
21:48:40 : 06/09
21:48:57 : 07/10
21:49:15 : 08/11
21:49:50 : 09/12
21:50:17 : 09/13
21:50:17 : Test finished.

----------
Total: 9/13
Probability that you were guessing: 13.3%

-- signature --
1f3af6709165e5766a497cd8e40f9bf838329f00

ABX test of frequency discrimination

Reply #24
Awesome, thanks.
"I hear it when I see it."

 
SimplePortal 1.0.0 RC1 © 2008-2019