Originally Posted by briansxx
Amir wrote: "That is certainly something to explore. But the more immediate question is, why did we in this thread and countless others assumed it was impossible to hear the differences? Didn't that bias you to think you couldn't pass the test initially?
How many people run around demanding "ABX DBTs" knew they could result in positive detection? I am pretty sure there would be far less forum noise and hostility if they knew they were making empty bluffs.
We do know what has changed though. People have learned how to listen. How to search for differences and reliably identify them. That claims of hearing damage are really excuses for not knowing how to listen critically.
The data we have, even without any investigation of cause, should if there is any logic to be had, completely change the landscape of these discussions. That there are people with better listening abilities so just because we can't hear a difference, doesn't mean others cannot. That professional training, just like any other field, matters. That just because you listen to music, it doesn't qualify you to reach conclusions regarding what others can hear.
We don't need any more data to reach these conclusions."
Amir, I believe the "bias" against the ability to hear differences arises from a number of elements, including: data about human hearing capabilities; Nyquist's theorem; published, controlled studies that indicate people cannot detect differences between "hi-res" and "redbook" versions of the same master at tolerable listening levels. For many posters here, including myself, the "no humanly-detectable differences" position seems both reasonable and supported by the available evidence. Your results challenge this "paradigm" and hence demand increased scrutiny.
If we look at the history of science, paradigm-challenging results fall into a number of categories, including (not an exhaustive list by any means):
- they are real and do refute the paradigm
- they arise from factors/artifacts that are spurious or do not challenge the paradigm
- they are specific to a very limited data set and not generalizable
- they are faked
So, in light of a surprising result, more data is indeed needed and the experiment needs to be repeated under controlled conditions and with many more samples to establish its 'reality." I agree that your result should change the landscape of these discussions, but not, perhaps, in the same ways you think it should.
Very good post Brian. Just a few points if I may
1. The theories you mention mean those things on paper/ideal execution. We can't build those systems. Bandlimiting with digital filters for example will mean ringing, or sacrifice of the pass-band response. Not saying these are audible necessarily but that they are considerations for real products as opposed to paper assumptions based on ideal math.
2. There is plenty of research that says 16 bits is not enough. Here is my earlier post from prosecution's witness, professor Vanderkooy:
Originally Posted by amirm
Let's review the latest argument. It started with a link to a blog/forum discussion with a person outlining what he heard at the local chapter of Audio Engineering Society:
Attended an AES seminar today given by John Vanderkooy of the University of
Waterloo, Waterloo Ontario entitled: "A Digital-Domain Listening Test for
Vanderkooy maintained that he has no doubt, after many ABX and DB tests
performed at the University that "...some people can, reliably, and in a
statistically meaningful way, detect the differences between 16-bit, 44.1 KHz
audio and the same program material recorded at 24-bit and 176.4 or 192 KHz."
He also added that there are also many people who CANNOT hear these
differences in a statistically meaningful way, so the jury is still out on
the efficacy or need for high-resolution digital audio.
Arny responded to the whole transcript with this:
I sent a copy of the above to John Vanderkooy, and his private emailed reply
included the word "misquote", and not in a good way. Suffice it to say that
the above in no way represents the talk that he thought he gave.
As we know in this thread, no record of Arny's email exchange remains with Professor Vanderkooy. So we are left with some detective work. For that, let's review the AES paper which Professor Vanderkooy had presented at Audio Engineering Society:
A Digital-Domain Listening Test for High-Resolution
Department of Physics and Astronomy, University of Waterloo, Waterloo, Ontario, Canada N2L 3G1
At the outset, let me state my bias that CD-quality audio
(44.1 kHz, 16 bit) is essentially transparent. Under
pristine conditions it may just be possible to hear
residual channel noise. Listening tests  have agreed
with this position, but there is much anecdotal evidence
that higher sampling rates and longer wordlengths have
significantly superior performance
I also noted but won't repeat here research papers in JAES from Stuart and separately from Fiedler, showing using math and listening test results of threshold of hearing that 16-bit channels have audible noise.
3. The listening tests you show are more flawed than any test we are doing here. The Meyer and Moran for example has no measurement whatsoever of the content they tested. Indeed it has turned out that they used bandlimited content. This is the kind of thing that would make Arny say "rookie mistake." Yet we rely on it like it is the bible.
The test also did not use trained listeners. Nor showed any attempt to find content that was revealing.
4. Following the last comment above, in these forums we had convinced ourselves that there is no such thing as a "golden ear" listener. In real life these people exist as well as sun rises from the east. And they get their label because they are able to perform listening tasks that blow ordinary people out of the water. And do so repeatedly across countless tests.
At the risk of appearing immodest, that is how I got that designation. An example was the search for a watermark for the DVD Audio format. Microsoft research had an algorithm they wanted to propose. Before the final submission I asked to test the samples to make sure the insertion of watermark was transparent. They gave me two versions of 3+ minute songs at 24/96 and asked me to find the mark.
The algorithm for watermark uses a perceptual model of human hearing and attempts to insert its bits where it would be masked by music. The number of bits inserted is a fraction of the total number of bits in the file. As such, the insertion algorithm will search and search and find the best spot to insert them. The insertion may change just a few PCM samples -- a needle in a haystack if there ever was.
Given the above, much of the track will then sound identical since nothing was inserted. At first I found the test daunting. I listened and listened and the files were identical. Then all of a sudden, I thought I heard the tiniest bit of distortion that I didn't recall was there before. So I proceeded to isolate that segment down to a fraction of a second. Then I was sure.
I communicated that to the researcher (really his manager) and he just about fell off his chair. They could not believe that I had managed to not only differentiate the files, but found the precise spot down to milliseconds. They looked at the code, found the bug and fixed it. No one else had found the difference before.
It is stories like above that repeated over and over again that gave me that reputation to be able to able to find distortions that others could not.
It is this kind of background that gives me a positive attitude as I enter these tests. I don't start with assuming it is impossible and give up as most people would.
None of this is known or talked about in these arguments on forums. Here is the worst part: when I tell these stories as I have to Arny for example, I just get ridicule and denial.
5. There are people who are better than me and some are untrained. Recall that my hearing sensitivity to high frequencies is shot so I can't hear distortions there as readily as others do.
Given all of this, I think the only thing unusual should be, after taking the test yourself, that our listening abilities sharply differ. A pathologist can read an X-ray a lot better than I can. It may seem like it is just a picture to look at but their training allows them to see the smallest anomalies which would go unnoticed by millions of ordinary people. This fact doesn't break any paradigm and should not in audio either.
Fact that others managed to also get there once given positive motivation shows that a lot of this is simply the wrong set of conclusions we have reached on these forums.
The best way then to scrutinize these results is to practice yourself. Knowing what you know now, see if you too can hear the difference. Once there, you can do all the analysis you want on yourself