Blind Test: Red Book vs. High Rez up to 352.8kHz/24bit

[sonitus mirus]

 
what greg said. a multimeter, a sonometer, or a microphone. it's something many people own already even if it's in the form of a cellphone with a crappy app. the precision might not be better than 0.1db, but it's already such an improvement compared to setting things by ear. and you just use some test tone as reference when it's about testing DACs.
 
for file format, a different thing. as we all recommend very strongly to convert a highres file to lower resolution yourself instead of trying to use what's available online in different formats(because it could be another master), there should not be any loudness difference if done correctly. like not adding replay gain to the conversion or stuff like that. as I've done a few times like a noob, using preset conversion routines^_^.
and the loudest part of signal is always, as mentioned, 0db. what you increase with 24bit is dynamic range below the 16bit of CD, not withing, not above. it lets us record silence in a much higher resolution. and 24bit format is increased dynamic range for the container of the music. the music itself rarely uses more than 70db of dynamic, be it on CD or highres.

 
I would also suggest that the converted, lower resolution file be upsampled back to match the format of the higher resolution file to further remove the possibility of the DAC handling various sampling rates differently.  This goes especially with higher priced boutique DACs with multiple filters and ludicrous specifications, or the Pono player.

 

[Baxide]

   
I would also suggest that the converted, lower resolution file be upsampled back to match the format of the higher resolution file to further remove the possibility of the DAC handling various sampling rates differently.  This goes especially with higher priced boutique DACs with multiple filters and ludicrous specifications, or the Pono player.

That what is lost during the downsampling of the file will not be recovered by upsampling again.

 

[Baxide]

 
what greg said. a multimeter, a sonometer, or a microphone. it's something many people own already even if it's in the form of a cellphone with a crappy app. the precision might not be better than 0.1db, but it's already such an improvement compared to setting things by ear. and you just use some test tone as reference when it's about testing DACs.
 
for file format, a different thing. as we all recommend very strongly to convert a highres file to lower resolution yourself instead of trying to use what's available online in different formats(because it could be another master), there should not be any loudness difference if done correctly. like not adding replay gain to the conversion or stuff like that. as I've done a few times like a noob, using preset conversion routines^_^.
and the loudest part of signal is always, as mentioned, 0db. what you increase with 24bit is dynamic range below the 16bit of CD, not withing, not above. it lets us record silence in a much higher resolution. and 24bit format is increased dynamic range for the container of the music. the music itself rarely uses more than 70db of dynamic, be it on CD or highres.

It won't work. The frequency response of the mic would give a gain level that is uneven even over the audio frequency in a 0dB sweep. Only white noise would offer you any sort of assurance. Using the actual audio file would give false results.

 

[castleofargh]

 

 
what greg said. a multimeter, a sonometer, or a microphone. it's something many people own already even if it's in the form of a cellphone with a crappy app. the precision might not be better than 0.1db, but it's already such an improvement compared to setting things by ear. and you just use some test tone as reference when it's about testing DACs.
 
for file format, a different thing. as we all recommend very strongly to convert a highres file to lower resolution yourself instead of trying to use what's available online in different formats(because it could be another master), there should not be any loudness difference if done correctly. like not adding replay gain to the conversion or stuff like that. as I've done a few times like a noob, using preset conversion routines^_^.
and the loudest part of signal is always, as mentioned, 0db. what you increase with 24bit is dynamic range below the 16bit of CD, not withing, not above. it lets us record silence in a much higher resolution. and 24bit format is increased dynamic range for the container of the music. the music itself rarely uses more than 70db of dynamic, be it on CD or highres.

It won't work. The frequency response of the mic would give a gain level that is uneven even over the audio frequency in a 0dB sweep. Only white noise would offer you any sort of assurance. Using the actual audio file would give false results.

bold part?
we don't care about the frequency response of the mic(at least not to volume match 2 devices), as long as it can repeatably give the same response when fed with the same signal, it's as good a tool as any.
noise fluctuates all the time in amplitude, it still could work but then you need some RTA that keeps track of the maximum levels(and at this point even a song could do).

 

[sonitus mirus]

 
That what is lost during the downsampling of the file will not be recovered by upsampling again.

 
Yes, but that misses my point.
 
Nothing has to be audibly changed when downsampling.  There are free tools available that can do this job much better, where I seriously doubt that anything approaching a 95% identity rate could be achieved by even the most critically trained listeners.
 
Normally a comparison would go like this:
 

• Convert original File A (24/192) file to 16/44.1: data will be lost in this process
• Convert 16/44.1 back to 24/192 as File B: the file will be limited to only the audible data from 16/44.1
• ABX the volume-matched File A and File B 

 
If the listener can identify a difference, then we can attempt to determine what might be the cause of this difference.
 
If we are fortunate enough to have a version from the same master with different sampling rates, as was apparently the case in the test in the previous link, then it could be interpreted that the downsampling process used was creating this difference, as the same listeners were unable to identify differences without the downsampling.
 
Why the testers did not attempt to try and downsample the files to be audibly the same is my question.  They had all the tools in place to insure that they were accurately downsampling the file, but they just ignored this step.  
 
In this case there was additional data to draw upon, but in other tests the game is sometimes rigged, and probably intentionally.  We've seen Bob Stuart's team use a known, inferior dithering technique in another recent test that created controversy, but we know the agenda he is promoting, so I wasn't surprised about that one.
 
https://secure.aes.org/forum/pubs/conventions/?ID=416
 
What if there were not two different versions of the same master?  It would have appeared that the 16/44.1 file was different, and probably inferior.  Some might debate that the issue is with the downsampling process, others might suggest it is the DAC and how it handles 44.1 vs 192.  And, of course, countless numbers of audiophiles would gush about how much better HD files are than Red Book, and how this test proves this once and for all.
 
If they could have established a proper downsampling technique that none of their listeners could accurately identify in an ABX test with the original HD file, the results would have been more valuable to me.

 

[spruce music]

   
Yes, but that misses my point.
 
Nothing has to be audibly changed when downsampling.  There are free tools available that can do this job much better, where I seriously doubt that anything approaching a 95% identity rate could be achieved by even the most critically trained listeners.
 
Normally a comparison would go like this:
 

• Convert original File A (24/192) file to 16/44.1: data will be lost in this process
• Convert 16/44.1 back to 24/192 as File B: the file will be limited to only the audible data from 16/44.1
• ABX the volume-matched File A and File B 

 
If the listener can identify a difference, then we can attempt to determine what might be the cause of this difference.
 
If we are fortunate enough to have a version from the same master with different sampling rates, as was apparently the case in the test in the previous link, then it could be interpreted that the downsampling process used was creating this difference, as the same listeners were unable to identify differences without the downsampling.
 
Why the testers did not attempt to try and downsample the files to be audibly the same is my question.  They had all the tools in place to insure that they were accurately downsampling the file, but they just ignored this step.  
 
In this case there was additional data to draw upon, but in other tests the game is sometimes rigged, and probably intentionally.  We've seen Bob Stuart's team use a known, inferior dithering technique in another recent test that created controversy, but we know the agenda he is promoting, so I wasn't surprised about that one.
 
https://secure.aes.org/forum/pubs/conventions/?ID=416
 
What if there were not two different versions of the same master?  It would have appeared that the 16/44.1 file was different, and probably inferior.  Some might debate that the issue is with the downsampling process, others might suggest it is the DAC and how it handles 44.1 vs 192.  And, of course, countless numbers of audiophiles would gush about how much better HD files are than Red Book, and how this test proves this once and for all.
 
If they could have established a proper downsampling technique that none of their listeners could accurately identify in an ABX test with the original HD file, the results would have been more valuable to me.

In the test I linked, where the 44 and 88 were not heard as different, but the downsample was I have a theory.  Until recently, like oh two or three years ago, virtually all downsampling software lowered the sound level of the downsampled file by .2 or .3 db.  This is just about where you begin getting positive differences in blind testing from level differences alone.  With good listeners and pretty good equipment that slight level difference may be all that was going on.  No mention was made of touching up the volume to compensate for that.  Many downsamplers still do this.  The reason is to prevent accidental digital overs when you change sample rates.  Without this or some other compensation it is possible to resample something and get some samples that are clipped even if the original file wasn't clipped anywhere.   I believe that test was done in 2010 so it is a pretty good bet that was the case.

 

[sonitus mirus]

 
In the test I linked, where the 44 and 88 were not heard as different, but the downsample was I have a theory.  Until recently, like oh two or three years ago, virtually all downsampling software lowered the sound level of the downsampled file by .2 or .3 db.  This is just about where you begin getting positive differences in blind testing from level differences alone.  With good listeners and pretty good equipment that slight level difference may be all that was going on.  No mention was made of touching up the volume to compensate for that.  Many downsamplers still do this.  The reason is to prevent accidental digital overs when you change sample rates.  Without this or some other compensation it is possible to resample something and get some samples that are clipped even if the original file wasn't clipped anywhere.   I believe that test was done in 2010 so it is a pretty good bet that was the case.

 
I would think that converting from 88 to 44 would be relatively trivial, even back in 2010.  Going from 96 or 192 to 44 would be trickier.  My thought is that the file was only passed once scanning for peaks, and rather than lower the volume level to eliminate any potential clipping, there was probably some audible clipping that could be heard after the sampling conversion.   Also, 64-bit, which may not have been available for this test, can now be regularly used when converting sampling rates to process calculations, which should theoretically offer higher precision, though I doubt it would make an audible difference.  Just guessing, but it seems more likely than volume differences or dither.  95% is rather obvious, to some extent, as I know just how difficult it can be to identify small differences in audio from my own testing.

 

[spruce music]

   
I would think that converting from 88 to 44 would be relatively trivial, even back in 2010.  Going from 96 or 192 to 44 would be trickier.  My thought is that the file was only passed once scanning for peaks, and rather than lower the volume level to eliminate any potential clipping, there was probably some audible clipping that could be heard after the sampling conversion.   Also, 64-bit, which may not have been available for this test, can now be regularly used when converting sampling rates to process calculations, which should theoretically offer higher precision, though I doubt it would make an audible difference.  Just guessing, but it seems more likely than volume differences or dither.  95% is rather obvious, to some extent, as I know just how difficult it can be to identify small differences in audio from my own testing.

There is no guessing to it.  Software whenever it downsampled dropped .2 to 3 db.  Some .2, some .25, and some .3 depending on whose software you used.  Even iZotope did that until not too long ago.  Currently iZotope, and Sox have fixed this issue.  As for other downsamplers I don't know.  I believe in 2010 all of them still dropped the level.  In the interest of speed I don't think any of them scanned the file for possible issues.  They simply had developed an algorithm and knew if you drop .25 db you will not get inter-sample overs.  The exact amount varying with the algorithm. Other than blind testing this would be a trival non-issue.  So they dropped level that amount and proceeded with the conversion.  Even if a file had nothing remotely close to 0 db this still happened upon conversion.
 
I don't know this was the cause of the down sampled file being audible, but it is possible.  Especially if they used one that dropped it .3 db. 

 

[castleofargh]

 I always assumed that the volume differences from playing different resolutions came from the DAC playing different sample rates instead of the files themselves.
I noticed that I could get slightly different values of replay gain with downsampled files compared to the highres one, but just like the stuff with the DR plug in I assumed it was some mess going on about phase or the cut ultrasounds showing a difference that had no meaning in the audible range.

 

[spruce music]

   I always assumed that the volume differences from playing different resolutions came from the DAC playing different sample rates instead of the files themselves.
I noticed that I could get slightly different values of replay gain with downsampled files compared to the highres one, but just like the stuff with the DR plug in I assumed it was some mess going on about phase or the cut ultrasounds showing a difference that had no meaning in the audible range.

Well if one had a file with extremely high ultrasonic energy you would remove it upon down-sampling which would let you raise the volume of the file.  Or leave it the same.  But that wasn't why the difference.  Re-sampling up or down can cause digital clipping not in the original.  So for instance before iZotope improved theirs it always dropped the resampled file .25 db in level.  I think Sox dropped it .2 db going from memory and many others drop it .3 db. I don't know how they fixed this, it may be they simply drop it enough to prevent clipping then change it back to match when done resampling or they may be doing something more sophisticated so this doesn't need to happen. 
 
With some resamplers I have access to I have sample rate converted a simple 1 khz sine wave.  Makes it easy to see how much they drop the signal level of if they keep it the same.

 

[castleofargh]

indeed resampling a single tone is pretty straightforward. yet I never thought about trying it.

I feel very silly right now.

 

[Dr4Bob]

I listen almost exclusively to lossless Redbook rips of my modest CD collection since going to computer files instead of spinning the digital platters. I recently obtained a variety of hi resolution tracks from a friend to evaluate. Whilst just my own impressions, I found matching 16/44 tracks ripped from CDs in ALAC compared to the same tracks in 24/96 and 24/192 in FLAC using Amarra for playback to be virtually identical. I must say I was somewhat surprised with all the hype over hi-rez; I am not suggesting those cursed with "golden ears" do not hear the difference but at this point in my own limited experience there is no real difference and I suspect we are again seeing the familiar point of diminishing returns in audio at best, another marketing technique at worst.

 

[Ancipital]

  I listen almost exclusively to lossless Redbook rips of my modest CD collection since going to computer files instead of spinning the digital platters. I recently obtained a variety of hi resolution tracks from a friend to evaluate. Whilst just my own impressions, I found matching 16/44 tracks ripped from CDs in ALAC compared to the same tracks in 24/96 and 24/192 in FLAC using Amarra for playback to be virtually identical. I must say I was somewhat surprised with all the hype over hi-rez; I am not suggesting those cursed with "golden ears" do not hear the difference but at this point in my own limited experience there is no real difference and I suspect we are again seeing the familiar point of diminishing returns in audio at best, another marketing technique at worst.

 
Worse, the higher res stuff is theoretically slightly counter-productive due to things like intermodulation distortion.
 
When the redbook standards were designed, those guys knew what they were doing. The sample rate gives a Nyquist frequency that fits human hearing very well, and the bit depth gives more than enough dynamic range to exceed it, too. 
 
High rate/depth audio is fantastic for the production process, but not only does it have no advantages for enduser listening, in some ways, it's actually a disadvantage. However, audiophile woowoo merchants seldom care about sampling theory or the biological limits of our hearing- if they can spend a borderline obscene amount to feel superior, then they will derive a lot of satisfaction from doing so.
 
Once the chin stroking starts, it all gets a bit...
 

 

[sonitus mirus]

Yep, if someone has tweeters that reach bat-like sonic levels, and they play 24/192 files, they might actually be introducing IMD that is negatively impacting the quality of the music.  The same system would be ok with standard 16/44.1 music, as the potential IMD would not be present.  
 
I'm sure that there is a better format, generally speaking, but what we already have is fantastic, and for all practical reasons it is as good as we can expect.  Personally, I don't believe it is worth the effort to attempt to find some better solution over what currently exists in Red Book, but with the digital age, I'd be content with a standardized 24/48 file going forward.  The major labels could simply convert everything to this format without having to remaster their libraries.  Just settle on a standard format and make everything available to purchase from a multitude of capable vendors. 
 
Sadly, the media conglomerates will most likely ruin any chance of a successful transition to the digital age.  We should all be able to purchase bit-perfect music from a vast majority of the catalogs at a similar price that CDs can be bought at Amazon and other retailers; but, so far, our options are either lossy mp3 or aac files, or ridiculously priced HD files. 

 

[icebear]

I recently got this CD (regular RB standard, not SACD):

 
And although I have no clue why some piano recorded by RVG sounds like a old dusty upright from a Westen movie but this remaster of the original analog tapes and formatted in lowly 16/44.1 sounds just freaking amazing. Most likely my jaw dropped to the floor but I don't remember, as I was so fascinated by this authentic session. Redbook is plenty sufficient in my book.

 
Getting the recording medium set up to run at it's best and placing the right microphones in the best position and get the musicians motivated to to have some fun playing music is about 100x times more important than 44.1, 88, 192 or 384khz DSD PCM MQA format choice. These recordings have been made in the 60's.
 
Taking painful care of all production steps seems to, here I mean the actual manufacturing to get the digital file onto the physical carrier, obviously needs attention. A few JVC K2 redbook CD's sound way better than the standard versions. The new formats are just being pushed to sell more hardware in the first place and then sell the coded high rez software files at 3 times the red book CD price that make the appropriate sample rate indicator light up. Exact same pattern with the latest and greatest MQA