Yes, well, you remember my test re-recording thru fruit, veggies and mud.
Only the noise floor gave it away, and then bot to many people.
I sure do
It was a real eye (ear) opener.
Till now, there has not been statistically valid proof that a listener is able to discern these existing files in the ABX test.
Based on nature of distortion measured (compare distortion with no load and 4ohm loaded) I claim that crossover distortion prevails all other distortions in the recording chain. Unless there are statistically valid positive results of this test, I see no reason to prepare another one on the same subject.
With all respect, I cannot count one isolated 7/8 caught on a single drum hit as a proof of audibility of the crossover distortion in this test.
Attachments
I would agree that the audibility of crossover distortion has not been proved although there does seem a possibility to discern audible 'differences'. Is that down to crossover distortion at HF on the shimmer of percussion... we don't know. I achieved a 7/8, a 6/8 and another 6/8. There were not lots of discarded attempts in between.
If it is purely crossover distortion that is of interest then I really think the set up should be identical for both files, with the amplifier in question biased to have more or less distortion. All other things are then equal.
Your files are impeccable as always, it is just the methodology that is suspect here imo.
If it is purely crossover distortion that is of interest then I really think the set up should be identical for both files, with the amplifier in question biased to have more or less distortion. All other things are then equal.
Your files are impeccable as always, it is just the methodology that is suspect here imo.
Yes, less than 1%, something like 11/12 at least and supported by complete text protocol that is generated by Foobar after every ABX test. Such protocol has a digital signature and can be checked if it is valid.
I am also not very happy by catching a short single drum hit. The files are dissimilar, the samples are not perfectly time aligned because it is a rip against recorded data (the smallest, zero time error is in the file beginning, getting worse when we approach end of files, making about 22 samples there). If we catch a single and non-musical short hit, the "distortion" might have occurred not only in the file itself, but also in a playback D/A chain of the testing person.
Last edited:
I am sorry, because something seems to go very wrong with this foobar. I have just made the test with 11/12 result. First 4 trials I did by listening, very seriously. Then I just continued by clicking X, X, X ... without any listening, until I reached 12 attempts. The result is 11/12 and the protocol is valid ...
Something is going very wrong.
Code:
foo_abx 2.0.2 report
foobar2000 v1.3.7
2017-11-03 13:28:01
File A: cc.wav
SHA1: 4f7cea25c5c93af637dc3dd7ad416402aa40eac4
File B: oo.wav
SHA1: eb0584e9e01746a0ee00ef60decf8eaca5832fcb
Used DSPs:
Resampler (PPHS)
Output:
WASAPI (event) : Speaker (USB Sound Blaster HD), 24-bit
Crossfading: NO
13:28:01 : Test started.
13:30:00 : 01/01
13:30:17 : 02/02
13:30:34 : 03/03
13:30:59 : 04/04
13:31:01 : 05/05
13:31:03 : 06/06
13:31:04 : 06/07
13:31:06 : 07/08
13:31:08 : 08/09
13:31:09 : 09/10
13:31:10 : 10/11
13:31:13 : 11/12
13:31:13 : Test finished.
----------
Total: 11/12
Probability that you were guessing: 0.3%
-- signature --
3bb6ce0ca03becfeda5f0864820c400539af5c98
Four runs where I just clicked either randomly or clicked the same choice, all gave expected results 4/8, 5/8 4/8 and a 5/8.
No answer to that one Pavel other than the fact that it could happen (such is the nature of statistics) but its unlikely.
You remember the infamous cap test. I used an on-line random generator to decide the sequence and was surprised it came out as something like X,X,X,X,Y,X,Y,Y,X. I mentioned that to SY at the time and he reckoned it was a valid result and still random.
No answer to that one Pavel other than the fact that it could happen (such is the nature of statistics) but its unlikely.
You remember the infamous cap test. I used an on-line random generator to decide the sequence and was surprised it came out as something like X,X,X,X,Y,X,Y,Y,X. I mentioned that to SY at the time and he reckoned it was a valid result and still random.
Great, thanks, Pavel. And hopefully you guys can figure out what's going on with the ABX (easy enough to use two disparate files to make sure the software is behaving)
As far as bringing confidence into an ABX, a larger n helps tremendously:
ABX Binomial Probability Table
What this table shows is how often you would stumble across at least this many answers by chance. E.g. Getting >= 6/8 correct would happen 14.5% of the time (with random guessing), but 12/16 only 3.8% of the time, and 15/20 only 2.1% of the time.
So either people need to be honest and aggregate multiple shorter runs (to catch "bad" sessions and "good"), or, better yet, crank up that trials per test to something like 20-30.
*This is entirely before adding in the disparate methods people are using to listen to these files.
Well, Pavel, it simply can happen. Purely statistically, you can (simply by chance!) hit the "correct" result while ABXing, even without listening. It would be rather unlikely to repeat that result, tho... Nothing to do with foobar - only after reasonably big number of trials you will get the "real" result. Try to toss a coin, you will see... A díky za boj proti high-endovým pověrám...
Michal
I'll be slightly provocative without intention to be impolite: generally, is ABXing the correct method? If you take one single file, make a copy with different name and let audience in good will make the ABX test, you will force results reflecting individual/psychological response to test method only - not the quality of the file itself. If you study how listening (as every individual cognitive process) works, you will recognize how difficult it is to get "objective" results from subjective process. Even more complicated is to interpret the results correctly...;-)
It could be that ABX-ing many trials with two very similar files teaches the brain to recognize them as the same. Maybe something useful for learning how to more reliably understand speech spoken a little differently each time. As usual, this kind of thing probably hasn't been studied and people are free to believe ABX is great, or greatly flawed, as they prefer.
My personal opinion is that we tend to use ABX because it's what we have, not because we know for a fact it's the best for a particular purpose. Maybe in some ways like using THD because its easy to measure, although in the case of THD, it has been researched enough to know a fair amount about its limitations.
My personal opinion is that we tend to use ABX because it's what we have, not because we know for a fact it's the best for a particular purpose. Maybe in some ways like using THD because its easy to measure, although in the case of THD, it has been researched enough to know a fair amount about its limitations.
I have already commented before, the difference is very appreciable, it is impossible to be wrong.
I suspect that the ABX comparator does some kind of processing with the signal that eliminates the difference.
In my systems I have always been able to differentiate between MP3 vs FLAC or 16/44 vs 24/96.
I suspect that the ABX comparator does some kind of processing with the signal that eliminates the difference.
In my systems I have always been able to differentiate between MP3 vs FLAC or 16/44 vs 24/96.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.