Apodizing filter

[Ghoostknight]

Ghoost, I am waiting for the thread where you talk about how the tin foil shield near your 4G router made your audio better.

As has been said, you seem to believe anything at all can affect audio so you implement something then listen for a change in audio quality, you hear it only because you are listening for it, then come up with a theory about why it makes a difference even when in reality the difference is probably only imagined.

You can get away with that in the tweaks threads but you can't expect not to be called out on it here.

I actually admire your enthusiasm and stickability but tempering that with a little common sense would be a good thing.

really no need for mambo jamboing around the actual point here

since i even can hear a 24khz lowpass in a DBT the psychosis must have went overboard...

 

[BS5711]

You do you then .......

 

[castleofargh]

really no need for mambo jamboing around the actual point here

since i even can hear a 24khz lowpass in a DBT the psychosis must have went overboard...

Again, if true, you either have wonderful young ears, or you play sounds much too loud for your ears, or your system doesn't do great with ultrasonic content.
I would do some measurements and find out what the problem is and if it's best to replace something or to avoid playing certain sample rates on that rig.

 

[gregorio]

well you can think that but i dont see a major flaw with my two recent DBT so far (and some others i posted some time ago)

I know you don’t. We’re going round in circles, me saying you simply ignore or don’t see the facts and you saying “but I don’t see”. I know you “don’t see” and worse still, you refuse to “see” even when the facts are explained to you and they are readily available on Wikipedia! So you’re agreeing with me but don’t even realise it, you’re simply blind to anything that doesn’t conform to the BS you’ve made-up!

where you simply concluded i faked them to keep your worldview intact

So again, you simply make-up BS to fill the gaps in your knowledge, which clearly are huge. The actual fact is that I concluded you probably just made a mistake, NOT that you faked the results, even though that is a possibility. I even listed some of the mistakes/errors you could have made but yet again (yawn) you just make up BS to fit your narrative and ignore/don’t see anything that contradicts it.

since i even can hear a 24khz lowpass in a DBT the psychosis must have went overboard...

Yes, we know you’re a superhuman and can hear way higher than other humans. Personally, if I were going to lie about being a superhuman I’d claim I’m faster than a speeding bullet, have a “spider sense”, be “un-killable” or have a more impressive superhero ability than just hearing 6-8kHz higher than other adults. You won’t get asked to join the Avengers with such a pathetic superpower! lol

G

 

[VNandor]

2. done multiple DBT on mp3 vs lossless so far, unless its a very bad codec i cant reliable get 100% of the answers right but overall i have done multiple tests now completing the "confidence" parameter above 95%, imo its highly unlikely that the 5% chance to get this purely by guesses were actually the reason why i "completed" multiple tests now, for me this is more of a hint that im "potentially" hearing a difference, which my subjective tests definitely confirm

well you can think that but i dont see a major flaw with my two recent DBT so far (and some others i posted some time ago) where you simply concluded i faked them to keep your worldview intact

the real question is how many blindtest i have to make (with >95% confidence) until we can conclude its not by guessing lol, tho my guess is hardcore objectivist have to either expierence it themself to finally believe it or conduct the test, mambo jamboing the DBT with additional variables till its probably 50/50 again

You are not doing the ABX test correctly. There is a problem with relying on feedback during the test. The problem is not that you check your guess after each answer, the problem is that you change what you do based on what you see and stop right at the result you want to see, effectively cherry picking the test. If you test this way, you can just keep going until the reported chance of not guessing is 95% which you can reach by relying on random guessing while waiting for a lucky streak and then you stop testing right after reaching 95% by pure chance.

An even worse way of cheating would be to do "some" undetermined amount of tests while getting "feedback" about the confidence, and reporting the highest number as the figure of how likely you were not just guessing. If you do 50 tests and at some point you reach 20/30 but you end up with a 25/50 then the chance of you not guessing is not high just because you got a lucky streak at around 30 tests. The parts of the trial that says you were likely just guessing shouldn't be ignored just because that is not what you want to see.

Again, the problem isn't that you check your results, the problem is that based on this result, you decide when to stop. If you first decide that you do x amount of tests no matter what, then checking the result of each trial won't skew the result. As an example, the chance of getting a heads (or tails) streak during 100 trials is way higher than getting way more heads (or tails) right at the 100th trial. If you don't decide on how much trials you are doing, you are effectively doing the test the first way I described and because of that, the equation used by the developer for calculating how likely the tester guessed is not correct. This is not mambo jambo as you put it, it's how math and probabilities work.

I've made a script because I wanted to check how much this cheats the result. It effectively creates the outcome 500 trials using a random number generator and reports how likely the results were not chosen randomly after each trial. It uses the same formula that your abx tester does which is n choose k divided by 2^n summed together for each X=k. You can fill in the outcomes yourself and compare it to the tester.
The script highlights the rows where the formula reports that the guess was "likely not random "(>95%) and prints them on the console. The console also reports the highest chance of not guessing.
https://jsbin.com/lalativife/1/edit?js,console,output

All i have to do is run the js a couple of times to get not just one but streaks of 95% chance of not guessing. Almost every run the highest chance of not guessing is way above 70%. The outcome is created by a random number generator but despite this, it's still easy to get long streaks of what looks like the outcome of the trials weren't likely randomized.
It looks like if the ABX test is done the way you do it, it is fairly easy to get 95% confidence of not guessing just by picking something randomly but doing enough tests and stopping at just the right time.

Of course, there wouldn't be any discussion about conditional probabilities if you clearly heard some difference and didn't pick the wrong answer so often. Picking the correct answer 20 times out of 33 on average is still quite random, not deterministic which is what the outcome of the test should be if the difference is clearly heard.

 

[Ghoostknight]

You are not doing the ABX test correctly. There is a problem with relying on feedback during the test. The problem is not that you check your guess after each answer, the problem is that you change what you do based on what you see and stop right at the result you want to see, effectively cherry picking the test. If you test this way, you can just keep going until the reported chance of not guessing is 95% which you can reach by relying on random guessing while waiting for a lucky streak and then you stop testing right after reaching 95% by pure chance.

An even worse way of cheating would be to do "some" undetermined amount of tests while getting "feedback" about the confidence, and reporting the highest number as the figure of how likely you were not just guessing. If you do 50 tests and at some point you reach 20/30 but you end up with a 25/50 then the chance of you not guessing is not high just because you got a lucky streak at around 30 tests. The parts of the trial that says you were likely just guessing shouldn't be ignored just because that is not what you want to see.

Again, the problem isn't that you check your results, the problem is that based on this result, you decide when to stop. If you first decide that you do x amount of tests no matter what, then checking the result of each trial won't skew the result. As an example, the chance of getting a heads (or tails) streak during 100 trials is way higher than getting way more heads (or tails) right at the 100th trial. If you don't decide on how much trials you are doing, you are effectively doing the test the first way I described and because of that, the equation used by the developer for calculating how likely the tester guessed is not correct. This is not mambo jambo as you put it, it's how math and probabilities work.

I've made a script because I wanted to check how much this cheats the result. It effectively creates the outcome 500 trials using a random number generator and reports how likely the results were not chosen randomly after each trial. It uses the same formula that your abx tester does which is n choose k divided by 2^n summed together for each X=k. You can fill in the outcomes yourself and compare it to the tester.
The script highlights the rows where the formula reports that the guess was "likely not random "(>95%) and prints them on the console. The console also reports the highest chance of not guessing.
https://jsbin.com/lalativife/1/edit?js,console,output

All i have to do is run the js a couple of times to get not just one but streaks of 95% chance of not guessing. Almost every run the highest chance of not guessing is way above 70%. The outcome is created by a random number generator but despite this, it's still easy to get long streaks of what looks like the outcome of the trials weren't likely randomized.
It looks like if the ABX test is done the way you do it, it is fairly easy to get 95% confidence of not guessing just by picking something randomly but doing enough tests and stopping at just the right time.

Of course, there wouldn't be any discussion about conditional probabilities if you clearly heard some difference and didn't pick the wrong answer so often. Picking the correct answer 20 times out of 33 on average is still quite random, not deterministic which is what the outcome of the test should be if the difference is clearly heard.

Well atleast this reasoning i can somewhat understand

So what would you say if i done 5 tests 20 trials each and completed 2 (or 3) out of 5 tests above 95% confidence ?
i guess this would be the next step to get a more bulletproof result but still, i imagine the general conses wouldnt change around here it being "random"

after all i know how i done these tests, its not like i repeated the test 10 times to give you an "good result", and imo i have done far too many tests with these kind of results to be just lucky... imo its a tendency that shows up

 

[VNandor]

So what would you say if i done 5 tests 20 trials each and completed 2 (or 3) out of 5 tests above 95% confidence ?

To reach a 95% confidence, you would have to guess correctly at least 59 times out of 100. The chances of that happening is yet again, different than what you propose, which is to get 2 tests above 95% because that means getting 2 tests consisting of 20 trials to at least 15/20.

If you are willing to do 100 trials then do 100 trials the way it's intended and don't play around with how you group the tests because if you do, the cookie cutter formula for the confidence doesn't apply and I assume you can't derive the formula for your own test format.

This doesn't mean that you are physically not allowed to split up the 100 trials and take breaks so your ears don't tire, it just means you shouldn't cheat the tests by grouping them in a way you that is convenient to you, like taking out the most successful 20 consecutive trials out of the total of 100 trials.

Let's just assume you can reach the 59/100 or even a 70/100. While that would surprise me, the next question I would genuinely be curious about, how come you ever pick the wrong one (let alone 30 times) if there's a difference that you always seem to hear? I don't think you ever said that sometimes you don't notice a difference but a 70/100 would indicate that you are not even close to always hearing a difference. I've done my fair share of lossy compression testing and I looked at the results' of other people, I don't recall ever seeing an "inbetween" result, most likely because when people hear a difference they just pick the correct ones all the time.

 

[bigshot]

Quite frankly, after this massive display of bias, I wouldn’t trust any test results unless the test was conducted by someone else who knows how to properly do it.

 

[Ghoostknight]

To reach a 95% confidence, you would have to guess correctly at least 59 times out of 100. The chances of that happening is yet again, different than what you propose, which is to get 2 tests above 95% because that means getting 2 tests consisting of 20 trials to at least 15/20.

If you are willing to do 100 trials then do 100 trials the way it's intended and don't play around with how you group the tests because if you do, the cookie cutter formula for the confidence doesn't apply and I assume you can't derive the formula for your own test format.

you guys think so backwards lol

the point was reaching 95% on 20 trials multiple times because the chance of guessing is already only 5%, getting 2-3 5% hits in 5 trials is pretty low by guessing

on the other hand i wouldnt trust a 59/100 result

 

[Ghoostknight]

Quite frankly, after this massive display of bias, I wouldn’t trust any test results unless the test was conducted by someone else who knows how to properly do it.

there is really no point in doing more blindtests here if you only accept 10/10 results

 

[VNandor]

you guys think so backwards lol

the point was reaching 95% on 20 trials multiple times because the chance of guessing is already only 5%, getting 2-3 5% hits in 5 trials is pretty low by guessing

on the other hand i wouldnt trust a 59/100 result

I see. It's lovely that you don't trust the 95% figure, so the previous test where you could barely get to 95% (by cheating as well) is not proof of anything. All you would have to do to get a more convincing result is to pick the correct sample more often which should not be hard if you hear a difference.

The chances of getting exactly 59/100 correctly by guessing can be calculated by 100 choose 59 divided 2 to the power 100. In general, that is n choose k divided by 2 to the power of n, where n is the number of trials and k is the number of correct guesses.

The chances of getting at least 59/100 correctly can be found by using the above formula and summing the results together from k=59 to k=100.

Can you walk me through what the chances are of finishing 2 tests out of 5 consisting of 20 trials with at least a 14/20 by just guessing? It seems you feel like that is less likely to happen but that does not amount to anything. I'm sorry but knowing math is not thinking backwards.

 

[gregorio]

So what would you say if i done 5 tests 20 trials each and completed 2 (or 3) out of 5 tests above 95% confidence ?

It all depends on what you are claiming. If you’re claiming to have aced an ABX between a Beethoven Symphony and a Mozart Symphony, then I’d not have any cause to dispute that because the difference between the two is both logically/rationally explainable as well as easily objectively demonstrated to be audible. However, if you’re claiming to have aced an ABX between say a fast roll-off linear phase filter and a fast roll-off apodizing filter then that’s NOT logically/rationally explainable (because the artefacts/ringing is above the range of human hearing), has NOT been objectively demonstrated to be audible and in fact, the reliable scientific evidence demonstrates this difference doesn’t even register in the auditory cortex, let alone is audible. So if you claim to have aced an ABX that’s contrary to all this science, then that’s an “extraordinary claim” and as laid out by one of the C20th great scientists: “An extraordinary claim requires extraordinary evidence” and some guy on the internet claiming to have aced an ABX test is NOT “extraordinary evidence”! If you really can demonstrate something so extraordinary, then get yourself to a university and tested independently by a researcher that knows what they’re doing. You should have absolutely no problem finding someone to do this because scientists are crying out for subjects that can demonstrate something extraordinary and be the basis of a juicy paper to earn their PhD, secure tenure or obtain a significant research grant!

you guys think so backwards lol

YES, at last, hallelujah brother you’ve got it, compared to you we do indeed think backwards!

What we think and do is research the scientific facts/evidence, OBJECTIVELY verify that we can hear what we think we’re hearing and if our test results conflict with the scientific facts/evidence then we search for faults/biases in our testing methodology. Only after extensive searching, including investigations/advice from independent expert others, would we tentatively claim that we might have managed to differentiate something that theoretically should not be discernible. So you’re right, this is entirely “so backwards” compared to you because you start at the end of this long difficult process by claiming to discern something that is inaudible, then you work your way backwards to the beginning. And if that’s not bad enough, your backwards journey is additionally invalid because you rely on cherry-picking evidence, inventing explanations, using biased tests to defend your claim/conclusion and all the while dismissing unbiased tests and reliable evidence on the basis that it must be wrong because it contradicts your (invalid) claims/conclusions!!

So yes, us guys, science itself and pretty much any rational person with an understanding of science and the scientific method “think so backwards” compared to you. For you to not even realise this and argue about it in an actual science discussion forum, even after it’s been explained to you numerous times, is thoroughly astonishing!!

G

 

[Ghoostknight]

Can you walk me through what the chances are of finishing 2 tests out of 5 consisting of 20 trials with at least a 14/20 by just guessing? It seems you feel like that is less likely to happen but that does not amount to anything. I'm sorry but knowing math is not thinking backwards.

the chance to hit 95% confidence by guessing once in 5 trials would be around 23%, two times 5%, three times 1,2%
i hope i got that right

 

[bigshot]

there is really no point in doing more blindtests here if you only accept 10/10 results

I only accept tests conducted to find out the truth, not ones conducted specifically to prove an argument. I think you have an extraordinary level of bias and it’s coloring everything you do and say.

 

[VNandor]

the chance to hit 95% confidence by guessing once in 5 trials would be around 23%, two times 5%, three times 1,2%
i hope i got that right

I don't think so but it's hard to tell because there is no reasoning behind why you came up with these numbers.