[gregorio]

  This thing about creating worse quality by purpose to drive sales of higher (more expensive) specs is sadly not uncommon and could be something that will push hires sales.

 
That's the problem, or problems! Firstly, as I alluded, a worse quality version may only be worse in certain listening environments, in other common (probably far more common) use environments, that worse quality version is actually perceived as better quality. Secondly, a mastering engineer may not know the intention/agenda of their client, he/she might believe they're making a version of higher quality for those other common listening environments, while their client might intend using it to demonstrate the superiority of HiDef and Thirdly, even if the mastering engineer does know of a nefarious intention, it's very unlikely he/she would risk their livelihood by publicly revealing it.
 
This is, as you say, not uncommon or in fact a particularly new phenomena, it's been going on since pretty much the availability of "HiRes" to the public nearly 20 years ago.
 
G

 

[mackie1001]

Of the back of that MQA appreciation thread mentioned earlier I thought I'd grab a free trial of Tidal Hifi and have a play. Straight away it was blatantly obvious that the difference I was hearing was due to it being a totally different master and nothing to do with the magical format at all. I set my vanilla Realtek laptop audio device to 16/44.1 plugged some £50 Sony IEMs in and had a listen. Not high end, revealing kit. Difference was very apparent. The MQA version probably 3+ db louder for starters. So in short, people raving about MQA are listening to different tracks. If they're hobbling the CD quality stuff too then it's just a big con. If they're not and just facilitating good remasters being created then that has some merit but MQA are getting money for really not very much. I'd happily listen to a 16/48 version of the same remaster in a FLAC container and not have the price jacked up by a license fee.

 

[dimedrol]

I've heard wonderful 16/44 records and awful 24/192 records. If every record out there was as good as the best 16/44 records, there wouldn't even be the need for a higher resolution format. People listen to bad 16/44 records and think that we need to increase the resolution. But it's not the resolution they they don't like. Higher resolution formats will eventually replace 16/44 anyway, but that is unlikely to yield a boost in the sound quality, not with current level of audio equipment.

 

[Computerpro3]

I was playing around tonight and came across this thread and the 8 bit vs. 16 bit test.  http://www.audiocheck.net/blindtests_16vs8bit_NeilYoung.php

I can pass this repeatedly with 95%+ confidence but I absolutely cannot articulate why, and it's driving me nuts.  If I listen to the whole sample and try to focus on different things to figure it out, I get it wrong.  If I listen for a couple seconds and make a snap judgement "this one is better" I can reliably pass.  Best I can figure out is some kind of treble muddiness but I have no idea.  Anyone else have a similar experience? 

Focusrite 2i4 DAC
Asgard 2 amp
HD650 headphones

EDIT:  I can not pass the Gangnam Style one with more than 60% confidence.

 

[spruce music]

  I was playing around tonight and came across this thread and the 8 bit vs. 16 bit test.  http://www.audiocheck.net/blindtests_16vs8bit_NeilYoung.php

I can pass this repeatedly with 95%+ confidence but I absolutely cannot articulate why, and it's driving me nuts.  If I listen to the whole sample and try to focus on different things to figure it out, I get it wrong.  If I listen for a couple seconds and make a snap judgement "this one is better" I can reliably pass.  Best I can figure out is some kind of treble muddiness but I have no idea.  Anyone else have a similar experience? 

Focusrite 2i4 DAC
Asgard 2 amp
HD650 headphones

EDIT:  I can not pass the Gangnam Style one with more than 60% confidence.

You may have discovered for yourself the effect of echoic memory.  It is why quick switching and very short segments give the finest most discriminating blind test results. 
 
https://en.wikipedia.org/wiki/Echoic_memory
 
Echoic memory only lasts a few seconds.  No more than 15 and probably less.  The auditory part of your brain holds  a few seconds of raw sensory input in memory.  You can switch instantly and for a few seconds directly compare two full perceptions, the current one and the one in echoic memory.  After that the perception goes to medium term and then longer term memory.  When it leaves echoic memory compression of the full perception occurs.  It is something like longer term memory is an MP3 of the full actual perception. 
 
So you need to use something that lets you listen to 2 or 3 seconds, switch near instantly and hear the same 2-3 seconds.  I find doing this lets me get high confidence scores on things I can never perceive over say 30 seconds.  Like different resampling algorithms.  If the difference is small enough this is needed, I also most of the time can't figure out why.  Another thing people who claim blind tests don't work don't want to believe.  I can sometimes get reliable repeatable blind results after I am listening and thinking I am just not hearing anything.  Yet some part of my hearing does perceive a difference I subjectively cannot pick up on. 

 

[pinnahertz]

 
You may have discovered for yourself the effect of echoic memory.  It is why quick switching and very short segments give the finest most discriminating blind test results. 
 
https://en.wikipedia.org/wiki/Echoic_memory
 
Echoic memory only lasts a few seconds.  No more than 15 and probably less.  The auditory part of your brain holds  a few seconds of raw sensory input in memory.  You can switch instantly and for a few seconds directly compare two full perceptions, the current one and the one in echoic memory.  After that the perception goes to medium term and then longer term memory.  When it leaves echoic memory compression of the full perception occurs.  It is something like longer term memory is an MP3 of the full actual perception. 

David Clark, in his AES paper, "High-Resolution Subjective Testing Using a Double-Blind Comparator" outlines the design of his hardware comparator.  He designed for a uniform 50ms gap between choices so if X=A the A>X switch would not reveal itself because of the absence of a switching gap. I have one of his units. I've found that reducing the gap has a minor but noticeable improvement on test resolution so long as the switch delay between all choices remains consistent. Using other switching methods, such as a fast but seamless crossfade, can be even more revealing. However, 2 seconds of gap would reduce the test resolution substantially. 

So you need to use something that lets you listen to 2 or 3 seconds, switch near instantly and hear the same 2-3 seconds.  I find doing this lets me get high confidence scores on things I can never perceive over say 30 seconds.  Like different resampling algorithms.  If the difference is small enough this is needed, I also most of the time can't figure out why. 

The highest resolution in ABX comparisons is to exactly synchronize both samples and switch without restarting the sample. Switching can be performed at any point during a test sample of any length. Unfortunately, many software ABX comparators preclude this type of comparison, which unfortunately does limit their resolution.  I have only a hypothesis of why this is, which I believe may be because repeating a short sample segment is highly unnatural, and contributes to the pressure of the test.  Playing a continuous sample and switching at any point lets the listener hear the DUT in a natural context, reducing the clinical nature of the test, and I believe, improving accuracy.  Sorry, I have no data to back up my hypothesis. I mean to test it some day, if some day ever comes. 

Another thing people who claim blind tests don't work don't want to believe.  I can sometimes get reliable repeatable blind results after I am listening and thinking I am just not hearing anything.  Yet some part of my hearing does perceive a difference I subjectively cannot pick up on.   

Most objections to the ABX/DBT protocol relate to it being "high pressure", a decision is demanded, and comparison time is limited. While the first part of that is true for data collection purposes, in practice there is no time limit. In fact, Clark's ABX comparator was designed to retain its X selections for a very long time permitting ABX tests lasting months.
 
BTW, as you may notice from his AES paper's title, ABX/DBT is actually a subjective test, not an objective test. Objective testing would be measurement.

 

[spruce music]

  David Clark, in his AES paper, "High-Resolution Subjective Testing Using a Double-Blind Comparator" outlines the design of his hardware comparator.  He designed for a uniform 50ms gap between choices so if X=A the A>X switch would not reveal itself because of the absence of a switching gap. I have one of his units. I've found that reducing the gap has a minor but noticeable improvement on test resolution so long as the switch delay between all choices remains consistent. Using other switching methods, such as a fast but seamless crossfade, can be even more revealing. However, 2 seconds of gap would reduce the test resolution substantially. 
The highest resolution in ABX comparisons is to exactly synchronize both samples and switch without restarting the sample. Switching can be performed at any point during a test sample of any length. Unfortunately, many software ABX comparators preclude this type of comparison, which unfortunately does limit their resolution.  I have only a hypothesis of why this is, which I believe may be because repeating a short sample segment is highly unnatural, and contributes to the pressure of the test.  Playing a continuous sample and switching at any point lets the listener hear the DUT in a natural context, reducing the clinical nature of the test, and I believe, improving accuracy.  Sorry, I have no data to back up my hypothesis. I mean to test it some day, if some day ever comes. 
Most objections to the ABX/DBT protocol relate to it being "high pressure", a decision is demanded, and comparison time is limited. While the first part of that is true for data collection purposes, in practice there is no time limit. In fact, Clark's ABX comparator was designed to retain its X selections for a very long time permitting ABX tests lasting months.
 
BTW, as you may notice from his AES paper's title, ABX/DBT is actually a subjective test, not an objective test. Objective testing would be measurement.

 
I have found switching that repeats the segment works better for me rather than switching an ongoing bit of music.  And I never meant a 2 second delay as that reduces discriminating ability of the listener substantially. 
 
Yes the testing segments can go on as long as one wishes.  Everytime it has been tried shorter is better, shorter is much better, very short is the best.  So if someone wishes to convince themselves fine.  They are only going to convince themselves they heard something sighted and dismiss the test when blinded they don't.
 
I think there is some truth to the high pressure claim.  You are asking people to listen in a way they aren't comfortable with and in informal tests I believe audiophiles think it is a challenge to their audiophile manhood.  It is like anything you haven't done much.  With repetition and experience it is old hat.  In the case of ABX testing it is tedious rather than pressure filled.  So people need convincing to do it a bit with little or nothing on the line until they aren't uptight about it.  Then you can get good results.
 
It is a subjective test.  Though with brain scans and such it may be more than that.  Already some years back simple monitoring of auditory nerve activity has been possible for academic testing.  That is one of the things done to show bone conducted ultrasonics are heard though usually an octave lower is how they are perceived.

 

[PETEBULL]

Originally Posted by globobock 
 
This thing about creating worse quality by purpose to drive sales of higher (more expensive) specs is sadly not uncommon and could be something that will push hires sales.

1. Download Hi-Rez content
2. Downsample it into 16/44.1
3. Upload it into your regular player
4. ???
5. Profit !!!

 

[globobock]

  1. Download Hi-Rez content
2. Downsample it into 16/44.1
3. Upload it into your regular player
4. ???
5. Profit !!!

I have a Sony DAP (the small A17) which has a down-sampling feature. Basically, what changes is the loudness.

 

[globobock]

  I was playing around tonight and came across this thread and the 8 bit vs. 16 bit test.  http://www.audiocheck.net/blindtests_16vs8bit_NeilYoung.php

I can pass this repeatedly with 95%+ confidence but I absolutely cannot articulate why, and it's driving me nuts.  If I listen to the whole sample and try to focus on different things to figure it out, I get it wrong.  If I listen for a couple seconds and make a snap judgement "this one is better" I can reliably pass.  Best I can figure out is some kind of treble muddiness but I have no idea.  Anyone else have a similar experience? 

Focusrite 2i4 DAC
Asgard 2 amp
HD650 headphones

EDIT:  I can not pass the Gangnam Style one with more than 60% confidence.

CMIIW, but on 8bit against 16bit, the difference should only be on the loudness levels, right? Not on the details?

 

[PETEBULL]

 
I have a Sony DAP (the small A17) which has a down-sampling feature. Basically, what changes is the loudness.

You got it wrong. If your player supports hi-res, you have no reason to downsample unless it doesn't handle it well. I meant that you can downsample files through a converter and use on conventional 16/44.1 hardware without buying hi-res DAC. 

 

[VNandor]

As far as I know, if you downsample from 16bit to 8bit the only thing that will change is the noisefloor. It goes way up. If there's more difference it it means there were additional processing made.

CMIIW, but on 8bit against 16bit, the difference should only be on the loudness levels, right? Not on the details?

 

 

[castleofargh]

 

 
I have a Sony DAP (the small A17) which has a down-sampling feature. Basically, what changes is the loudness.

You got it wrong. If your player supports hi-res, you have no reason to downsample unless it doesn't handle it well. I meant that you can downsample files through a converter and use on conventional 16/44.1 hardware without buying hi-res DAC. 

the sony DAP doesn't allow to use DSPs with highres files. IDK if it's a processing matter or if the DSPs are simply working at a given rate, but that's how it goes on that DAP. so as a cheap alternative, sony has a downsapling option to "enable" DSPs even with those highres files. but of course at that point it makes more sense to import lower res files on the DAP from the start IMO.
now the loudness part is a mistake from our friend, the option is indeed a downsampling one, the bit depth and/or the loudness aren't affected.

 

[gregorio]

  CMIIW, but on 8bit against 16bit, the difference should only be on the loudness levels, right? Not on the details?

 

As far as I know, if you downsample from 16bit to 8bit the only thing that will change is the noisefloor. It goes way up. If there's more difference it it means there were additional processing made.

 
If doing a bit reduction (only!) then there should be no difference in loudness and VNandor is correct, the only difference will be a far higher noise floor, which would likely affect the amount of details you can hear. However, how noticeable that far higher noise floor would be and how much detail you therefore wouldn't hear would depend on the actual audio/music plus the amount/distribution of the dither noise-shaping. In practice, with some material and an appropriate noise-shaping algorithm, it might be quite difficult to tell apart an 8bit version from a 16bit one.
 
If converting from hires to say 16/44.1, then we're not ONLY doing a bit reduction, we're doing a resampling process as well. In this case there will often be a change in loudness (level) as a good sample rate converter should take into account inter-sample peaks and reduce the level to avoid clipping, typically these SRCs will only reduce the level by about 0.2dB or so but that can be enough in some cases to screw up a blind or DBT!
 
G

 

[Computerpro3]

CMIIW, but on 8bit against 16bit, the difference should only be on the loudness levels, right? Not on the details?

I have no idea what the technical difference between the two files is, or if the test is even valid.  All I know is that I can reliably pass it.  I cannot do the same for the Gangam Style test.