Conclusive "Proof" that higher resolution audio sounds different
- Thread starter amirm
- Start date
No, of course not. We don't even have a statistically significant sample of the test that started this thread, much less re-testing, controls, all the things that are required to even begin talking about data. We're a long way from proof. But if, even given all the weaknesses, you believe there is a difference (and I do), we have almost nothing to indicate what that difference might be. There were times it was an interesting discussion, though. And other times it was an infuriating mess.
Tim
I can conclusively say that is wrong.I think the only thing that we can all agree on is that nothing has been proven. Conclusively or otherwise
Of course we do Tim. The results of the testing I did had probability of chance being essentially zero. With respect to my hearing abilities running that test, with my equipment, that is statistical significance.
Do you mean something else with that statement?
Amir, I think a summation of what you think was proven, and why, would be goodI can conclusively say that is wrong.
Are we talking about the original test? The one in which you were the only subject? How many trials did you run? Maybe I'm confused.
Tim
Tim,
I warned you several times about the importance of statistics in these matters ... The results of Amir (or others) tests need proper statitical analysis to be interpreted and valuated. We are not debating the usual audiophile challenge carried between two drinks.
.
My hope, in asking for consensus, was to determine what folks could "reasonably" agree to (and move forward from there) and not restart this discussion from Post No. 1. Do we, as reasonable adults, really need to continue to regurgitate what has already been said ad nauseum?
But then again, maybe we'll set a new record on the number of posts made without resolving anything.
Think Washington politics.
But then again, maybe we'll set a new record on the number of posts made without resolving anything.
Think Washington politics.
Well that is the point partly Amir had in mind. There is quite the limit on what can be resolved with such things. Along with the fact many in general have jumped to a DBT and not dotted I's or crossed T's to make it even useful for them personally. That more attention needs paying to such matters. etc/ etc/
So what has been resolved? Amir, myself and a few others could using Foobar ABX pick out the original jangling keys file 44/16 vs 96/24. For myself resampling with Sox made it no longer possible. Amir resampled with Adobe Audition which is also a quality resampler though I don't know if it time aligns results the way Sox does. If not that is a point worth looking into as per Tony Lauck's cautions about switching artifacts. I would be very curious if Amir could detect 44/24 or 48/24 vs the original in jangling keys. As far as I remember he never did that comparison with Arny's files.
Beyond that lots of contentious debate about IMD which seems technically a non-issue though Arny I don't think agrees. If there is anything else accomplished beyond bickering I missed it.
So what has been resolved? Amir, myself and a few others could using Foobar ABX pick out the original jangling keys file 44/16 vs 96/24. For myself resampling with Sox made it no longer possible. Amir resampled with Adobe Audition which is also a quality resampler though I don't know if it time aligns results the way Sox does. If not that is a point worth looking into as per Tony Lauck's cautions about switching artifacts. I would be very curious if Amir could detect 44/24 or 48/24 vs the original in jangling keys. As far as I remember he never did that comparison with Arny's files.
Beyond that lots of contentious debate about IMD which seems technically a non-issue though Arny I don't think agrees. If there is anything else accomplished beyond bickering I missed it.
No, we don't need to rehash this monstrosity. Simple answer to your simple question, Dude: nothing has been resolved. I think most of us can agree that a difference was heard by a handful of people. We really don't have anything more than that.
Tim
Tim
And that is indeed significant. It had been widely proclaimed for the past 14 years that the files that have now been definitively distinguished by ear could not so distinguished under any circumstances. That they could not be distinguished "proved" that there was no audible advantage under any circumstances to sampling audio at greater than 44.1kHz with a bit depth greater than 16.
As listeners now have shown to a very high statistical significance that the two files sounded different, all that "proof" collapses. That has clearly been resolved.
The question now becomes why? I understand that that the answer to that question will be discussed in the Meridian paper being presented to next month's Audio Engineering Society convention.
John Atkinson
Editor, Stereophile
Yes, John, that is the interesting question but not much progress being made on it here in 1400 posts
I propose suspending all further discussion in this thread (unless of a technical nature) & reconvene when the Meridian paper is presented & published.
(I realise I'm partly responsible for a lot of the distraction from the technical focus that should have been the kernel of this thread & for that I apologise).
I reported on this many pages before. I passed one out of his three tests:
Here is the post with more info: http://www.whatsbestforum.com/showt...unds-different&p=279735&viewfull=1#post279735
Amir, I'm downloading the files right now and will give it a try. Do you have any idea about the spl you were monitoring at ? IME this is crucial for audibility of low level noise. Everyone has a hearing threshold
The problem though is that in reality the blog is NOT doing 16bit vs 24bit, they are doing native (although ideally should be in-depth analysis of the PCM files for validation before testing) 24bit vs decimated.
If they were actually doing this from the age old debate of 16bit vs 24bit then you would need to record twice at different native bit depths (and rates if that is also part of the context); once for 16bit and then once for 24bit.
The result would probably show that in reality 16bit is not enough (technically been shown ideally need roughly 20bits "normally" and 32bits if wanting to ensure-guarantee transparency transcoding of DSD -> PCM) when one also considers the studio and the process-chains involved.
So the debate needs to change the age old argument more towards; native 24bit and sampling rate vs downsampled/decimated/upsampled/transcoded (PCM<--->DSD).
It may seem I am being a bit pedantic but this is an important differentiation from what many still think of from the past POV on 16bit vs 24bit, that then helps define how one approaches the scope-focus-context.
Thanks
Orb
Last edited:
Unfortunately, there is no such thing as a "native" 24 bit analog to digital converter. Existing converters operate at a high sampling rate and low bit depth and then digital signal processing is employed to produce the final output format. From an experimental perspective, if one used the "same" converter and output in two different formats one would actually be using two different (firmware) algorithms. Under the circumstances, one might as well just take a 24 bit file and convert that to 16 bits via software. This is a reasonably straightforward process and one could even post all the relevant source code in a single WBF posting if one were so inclined.
Assuming "native" 24 or 16 bit converters were avalable, two separate converters would be needed, so that a live microphone feed could be used. Using a recording of any form (e.g. tape, or DAC playback) would negate the "native" aspects of the test. Two converters adds a confounding aspect, in that two "identical" converters might not actually be "identical". It might be possible to design a suitable experiment to take account of the confounding effects of any differences (e.g. to the analog circuitry in the two boxes), but this would complicate statistical analysis of experimental data.
In any event, the argument that the double conversion invalidates the experiment would only come into force to refute positive results. It would seem a much better approach to get these first and then look at exactly what conversion algorithm was used and try others. Anyone who has worked with bit depth conversions and tried different variants will realize that there are various alternatives and that they sound different, as can be readily observed by starting with a low level recording, doing the conversion, and then playing back the results at increased volume.
BTW, Archimago's test was at 96 Khz. A bit depth test run at 44.1 kHz will be easier to pass since noise and quantization distortion are spread out by higher sample rates.
Tony,
IMO that is taking it even further down the logical path and one I think that is too far (been discussed many times about what is a true PCM DAC or true DSD DAC/most are "not true" to PCM or DSD due to nature of the circuitry involved or integrated within the DAC, and to a certain extent wide-narrow editing)
If we take this as far as you do, then ANY test involving DACs must utilise multiple different DAC solutions, which at some point the investigation may need to go that far I appreciate (or look at filter coefficients/linear-minimum phase); but that is a very complex subject tbh as can be seen by the 70+ page thread discussions we have had on that in the past.
For now I cannot see the issue with native being defined outside of the DAC itself; meaning the original sampling rate and bit depth recorded at and maintained (lets assume the correct hardware and setup is used, and as I mentioned earlier the PCM or DSD files can be analysed and validated to ensure they have not been upsampled/downsampled-decimated/transcoded from one format/spec to another.
Otherwise you are just doing a 24bit decimation to 16bit test (along with whatever dither/algorithms used).
So unfortunately that also invalidates the test if focus is specifically 16bit vs 24bit - note this is a different distinction to what Amir is doing and not same as the blog; Amir is showing that he is picking up a difference between an original file (which would had original recording rate/bit depth) to when it is downsampled/decimated.
As an example; lets say there is something different that Amir is picking up, well it can only be related to downsampling-decimation-dither because we only have an original 24-bit file that must be processed to 16-bit; assuming everything else is ok in terms of distortion/volume/synch offset/etc.
And that is why the linked blog is wrong to say 16bit VS 24bit, because in reality it is just 24bit decimated to 16bit, we never started with an original 16bit native recording to compare to a reiterated recording at 24bit (appreciate this has much complexity to it but then it is about the scope-context one is trying to present).
Anyway we are digressing from my OP a bit.
Edit:
Please read next post as clarify why IMO this test setup is not really about 16bit vs 24bit.
Cheers
Orb
IMO that is taking it even further down the logical path and one I think that is too far (been discussed many times about what is a true PCM DAC or true DSD DAC/most are "not true" to PCM or DSD due to nature of the circuitry involved or integrated within the DAC, and to a certain extent wide-narrow editing)
If we take this as far as you do, then ANY test involving DACs must utilise multiple different DAC solutions, which at some point the investigation may need to go that far I appreciate (or look at filter coefficients/linear-minimum phase); but that is a very complex subject tbh as can be seen by the 70+ page thread discussions we have had on that in the past.
For now I cannot see the issue with native being defined outside of the DAC itself; meaning the original sampling rate and bit depth recorded at and maintained (lets assume the correct hardware and setup is used, and as I mentioned earlier the PCM or DSD files can be analysed and validated to ensure they have not been upsampled/downsampled-decimated/transcoded from one format/spec to another.
Otherwise you are just doing a 24bit decimation to 16bit test (along with whatever dither/algorithms used).
So unfortunately that also invalidates the test if focus is specifically 16bit vs 24bit
- note this is a different distinction to what Amir is doing and not same as the blog; Amir is showing that he is picking up a difference between an original file (which would had original recording rate/bit depth) to when it is downsampled/decimated.As an example; lets say there is something different that Amir is picking up, well it can only be related to downsampling-decimation-dither because we only have an original 24-bit file that must be processed to 16-bit; assuming everything else is ok in terms of distortion/volume/synch offset/etc.
And that is why the linked blog is wrong to say 16bit VS 24bit, because in reality it is just 24bit decimated to 16bit, we never started with an original 16bit native recording to compare to a reiterated recording at 24bit (appreciate this has much complexity to it but then it is about the scope-context one is trying to present).
Anyway we are digressing from my OP a bit.
Edit:
Please read next post as clarify why IMO this test setup is not really about 16bit vs 24bit.
Cheers
Orb
Last edited:
Just to add and clarify,
the 16bit in all these discussions with comparison to 24bit is a bit of a misnomer; because one in reality is NOT comparing 16bit recording to 24bit recording, one is comparing a 24bit music recording to how it is after being processed to 16bit.
So we are currently looking at only the transparency of upsampling/downsampling/decimation/transcoding.
Same way comparing DXD to DSD but starting with say original "native" DSD and then transcoding to DXD for ABX (although technically transcoding for DSD -> PCM is meant to require 32bits and not 24bits to be guaranteed if looking at Philips research); in reality you are not specifically comparing DSD and DXD but primarily the transcoding solution used and its transparency.
Appreciate here internal-integrated DAC solutions for a real test would also be a major consideration for controlled environment-system.
Thanks
Orb
the 16bit in all these discussions with comparison to 24bit is a bit of a misnomer; because one in reality is NOT comparing 16bit recording to 24bit recording, one is comparing a 24bit music recording to how it is after being processed to 16bit.
So we are currently looking at only the transparency of upsampling/downsampling/decimation/transcoding.
Same way comparing DXD to DSD but starting with say original "native" DSD and then transcoding to DXD for ABX (although technically transcoding for DSD -> PCM is meant to require 32bits and not 24bits to be guaranteed if looking at Philips research); in reality you are not specifically comparing DSD and DXD but primarily the transcoding solution used and its transparency.
Appreciate here internal-integrated DAC solutions for a real test would also be a major consideration for controlled environment-system.
Thanks
Orb
Similar threads
