INTERNET BLIND TEST: MQA Core Decoding vs. Standard Hi-Res (24/88 or 24/96)

Discussion in 'Audio Hardware' started by Archimago, Jul 15, 2017.

  1. Archimago

    Archimago Well-Known Member Thread Starter

    Hi everyone. If you're curious about MQA and want to have a listen for yourself the effect it makes / doesn't make, come check out a little blind test I'm conducting.

    Download the files, have a listen, complete my survey!

    Will be interested in what you, the listener perceives...

    INTERNET BLIND TEST: MQA Core Decoding vs. Standard Hi-Res Audio

    BTW: Test will close for survey entries on September 8th, 2017.

    Enjoy!
     
  2. avanti1960

    avanti1960 Forum Resident

    Location:
    Chicago metro, USA
    Intreresting test. I'm in. I'll be listening on my Sony HAP Z1ES later today.
     
  3. Tommy SB

    Tommy SB Forum Resident

    Location:
    Santa Barbara, CA
    Will give this a try later.

    FYI, track "Arnesen A" is showing a time of 1:54 vs. 1:53 for "Arnesen B" in foobar. Looks like "Arnesen A" just needs to be cropped a bit to avoid any visual cues.
     
    Last edited: Jul 15, 2017
  4. Tullman

    Tullman I prefer analog

    Location:
    Boston MA
    I Think I will give this a try as well.
     
  5. c-eling

    c-eling Forum Resident

    Done, thanks Arch :cheers:
     
  6. avanti1960

    avanti1960 Forum Resident

    Location:
    Chicago metro, USA
    test completed, definitely preferred one track over the other in all three cases but may have picked the same process only twice, which means one or the other sounds better depending on recording....
     
  7. missan

    missan Forum Resident

    Location:
    Stockholm
    No particular difference I could trust,
     
    c-eling likes this.
  8. c-eling

    c-eling Forum Resident

    I had issues with the third sample Missan. I couldn't distinguish between them.
    As a whole if this is it on what the process is going to sound like, it's going to be a huge pass for me, unless somehow they have access to a different master/ing.
     
  9. missan

    missan Forum Resident

    Location:
    Stockholm
    I don´t think I´m particularly good in hearing differences this way, but anyway, they sound rather similar to me.
     
    c-eling likes this.
  10. Archimago

    Archimago Well-Known Member Thread Starter

    Thanks, yeah, I didn't cut off the ends exactly and added a bit of fade out as well. I made sure to focus on the start of the track for alignment for all those who want to use ABX Comparator or similar software to flip back and forth.
     
  11. Archimago

    Archimago Well-Known Member Thread Starter

    Thanks, as perhaps expected, differences may be more noticeable on certain tracks and even portions of tracks depending on the complexity of the music.
     
  12. Archimago

    Archimago Well-Known Member Thread Starter

    In terms of *magnitude* of the "effect", IMO this is exactly what you should expect when listening to TIDAL Master/MQA and comparing it with the equivalent mastering from a HDtracks download.
     
    Robert C and c-eling like this.
  13. Archimago

    Archimago Well-Known Member Thread Starter

    Don't think... Feeeel. ;)
     
    bhazen, kBear, LarryP and 1 other person like this.
  14. missan

    missan Forum Resident

    Location:
    Stockholm
    I felt I could not point out which was which.
     
    bhazen, Pinknik and Gaslight like this.
  15. testikoff

    testikoff Forum Resident

    This is actually a great opportunity for listeners with "golden ears" to prove that they can reliably discern the differences between A & B versions. I failed to do so with excerpts of tracks 1 & 2 (perfectly lined up in time domain & with the exact same number of samples), which seem to exhibit the largest difference between them out of 3 pairs offered:
    Code:
    foo_abx 2.0.2 report
    foobar2000 v1.3.15
    2017-07-16 00:31:41
    
    File A: 01 - Arnesen A [excerpt].flac
    SHA1: b0c8c1ba746d3c4b3a456522f79ceaf28800afa4
    File B: 02 - Arnesen B [excerpt].flac
    SHA1: dda477c632cbbb3053880a332fb2f046fed985cb
    
    Output:
    ASIO : Meridian USB ASIO Driver
    Crossfading: NO
    
    00:31:41 : Test started.
    00:32:25 : 01/01
    00:32:53 : 02/02
    00:33:47 : 02/03
    00:34:01 : 03/04
    00:34:11 : 03/05
    00:34:26 : 04/06
    00:34:45 : 04/07
    00:34:56 : 04/08
    00:35:13 : 04/09
    00:35:38 : 04/10
    00:35:38 : Test finished.
    
     ----------
    Total: 4/10
    Probability that you were guessing: 82.8%
    
     -- signature --
    09916f4dd2cb3b8a8e6053ad1fb7b56244aa1608
    
    The guys preferring one track over the other should try distinguishing them in an honest non-sighted comparison. If they are indeed successful, then I can start taking their opinions seriously... ;)



    P.S.
    Here is the DR analysis of the null-test delta for excerpts of tracks 1 & 2 I used in my ABX test:
    Code:
    foobar2000 1.3.15 / Dynamic Range Meter 1.1.1
    log date: 2017-07-16 00:51:37
    
    --------------------------------------------------------------------------------
    Statistics for: ?-01 - Arnesen AB delta  [excerpt]
    Number of samples: 18506743
    Duration: 1:45
    --------------------------------------------------------------------------------
    
                     Left              Right
    
    Peak Value:    -40.02 dB   ---    -39.13 dB 
    Avg RMS:       -59.02 dB   ---    -58.55 dB 
    DR channel:     12.21 dB   ---     12.62 dB 
    --------------------------------------------------------------------------------
    
    Official DR Value: DR12
    
    Samplerate:        176400 Hz
    Channels:          2
    Bits per sample:   24
    Bitrate:           2928 kbps
    Codec:             FLAC
    
    Delta peak is at -39dBFS.
     
    Last edited: Jul 16, 2017
    Chooke, englishbob and Tommy SB like this.
  16. avanti1960

    avanti1960 Forum Resident

    Location:
    Chicago metro, USA
    I had no problem hearing the differences for all three samples. The differences are subtle but easily recognized on my system. On the first sample see if you can hear one track just a little warmer and slightly veiled than the other.
     
  17. testikoff

    testikoff Forum Resident

    A successful ABX test (i.e. 9+ successes out of 10 attempts) report?..
     
    Last edited: Jul 16, 2017
    Robert C likes this.
  18. avanti1960

    avanti1960 Forum Resident

    Location:
    Chicago metro, USA
    10/10. seriously.
     
  19. liv3evil

    liv3evil Forum Resident

    Location:
    NY USA
    Non-trolling question: doesn't this level of tinkering derail A/B comparison? Isn't it important to keep samples disparate, as opposed to (presumably) adding a fade in the same (audio) application? i.e. both samples have had their 1s and 0s modified by X application.
     
  20. testikoff

    testikoff Forum Resident

    No ABX report, no trust. Sorry...
     
  21. testikoff

    testikoff Forum Resident


    Here are the FB2K's bit comparison results for the original test samples you provided:


    Code:
    Differences found in compared tracks.
    Non-zero offset detected.
    
    Comparing:
    "J:\AUDIO\Comparison\MQA\01 - Arnesen A.flac"
    "J:\AUDIO\Comparison\MQA\02 - Arnesen B.flac"
    Differences found: length mismatch - 1:53.783231 vs 1:53.334031, 20071362 vs 19992123 samples.
    Compared 19992123 samples, discarded last 79239 samples from the longer file.
    Differences found within the compared range: 39984115 values, starting at 0:00.000006, peak: 1.1926856 at 1:09.906740, 1ch
    Channel difference peaks: 1.1926856 1.1923950
    File #1 peaks: 0.6952808 0.7752248
    File #2 peaks: 0.6952226 0.7754351
    Detected offset as 2147 samples.
    
    Comparing again with corrected offset...
    Differences found: length mismatch - 1:53.771060 vs 1:53.321859, 20069215 vs 19989976 samples.
    Compared 19989976 samples, with offset of 2147 discarding last/first samples from total of 19992123, discarded samples were not silent in either file, discarded last 79239 samples from the longer file.
    Differences found within the compared range: 39974143 values, starting at 0:00.000000, peak: 0.0110536 at 1:10.913509, 2ch
    Channel difference peaks: 0.0099723 0.0110536
    File #1 peaks: 0.6952808 0.7752248
    File #2 peaks: 0.6952226 0.7754351
    
    
    
    
    Differences found in compared tracks.
    Non-zero offset detected.
    
    Comparing:
    "J:\AUDIO\Comparison\MQA\03 - Gjeilo A.flac"
    "J:\AUDIO\Comparison\MQA\04 - Gjeilo B.flac"
    Compared 23040000 samples.
    Differences found: 46079769 values, starting at 0:00.000266, peak: 0.9052150 at 1:31.125411, 1ch
    Channel difference peaks: 0.9052150 0.8821592
    File #1 peaks: 0.5090258 0.7484424
    File #2 peaks: 0.5090773 0.7484064
    Detected offset as 160 samples.
    
    Comparing again with corrected offset...
    Compared 23039840 samples, with offset of 160 discarding last/first samples from total of 23040000, discarded samples were not silent in either file.
    Differences found within the compared range: 46002322 values, starting at 0:00.000000, peak: 0.0184191 at 1:58.453953, 1ch
    Channel difference peaks: 0.0184191 0.0152308
    File #1 peaks: 0.5090258 0.7484424
    File #2 peaks: 0.5090773 0.7484064
    
    
    
    
    Differences found in compared tracks.
    Non-zero offset detected.
    
    Comparing:
    "J:\AUDIO\Comparison\MQA\05 - Mozart A.flac"
    "J:\AUDIO\Comparison\MQA\06 - Mozart B.flac"
    Compared 21168000 samples.
    Differences found: 42335852 values, starting at 0:00.000000, peak: 1.6764789 at 0:34.090465, 1ch
    Channel difference peaks: 1.6764789 1.3210939
    File #1 peaks: 0.9116414 0.8876039
    File #2 peaks: 0.9112427 0.8873681
    Detected offset as -2318 samples.
    
    Comparing again with corrected offset...
    Compared 21165682 samples, with offset of -2318 discarding last/first samples from total of 21168000, discarded samples were not silent in either file.
    Differences found within the compared range: 42287136 values, starting at 0:00.000000, peak: 0.0655744 at 1:58.016627, 2ch
    Channel difference peaks: 0.0364155 0.0655744
    File #1 peaks: 0.9116414 0.8876039
    File #2 peaks: 0.9112427 0.8873681
    
    
    
    


    I made my own excerpts from your files perfectly lined up time-wise to do ABX + listening tests:
    Code:
    Differences found in compared tracks.
    Zero offset detected.
    
    Comparing:
    "J:\AUDIO\Comparison\MQA\01 - Arnesen A [excerpt].flac"
    "J:\AUDIO\Comparison\MQA\02 - Arnesen B [excerpt].flac"
    Compared 18506743 samples.
    Differences found: 37008466 values, starting at 0:00.000000, peak: 0.0110536 at 1:07.343243, 2ch
    Channel difference peaks: 0.0099723 0.0110536
    File #1 peaks: 0.6952808 0.7752248
    File #2 peaks: 0.6952226 0.7754351
    Detected offset as 0 samples.
    
    
    
    
    Differences found in compared tracks.
    Zero offset detected.
    
    Comparing:
    "J:\AUDIO\Comparison\MQA\03 - Gjeilo A [excerpt].flac"
    "J:\AUDIO\Comparison\MQA\04 - Gjeilo B [excerpt].flac"
    Compared 21285468 samples.
    Differences found: 42498552 values, starting at 0:00.000000, peak: 0.0016105 at 1:29.931010, 2ch
    Channel difference peaks: 0.0014882 0.0016105
    File #1 peaks: 0.5090258 0.7484424
    File #2 peaks: 0.5090773 0.7484064
    Detected offset as 0 samples.
    
    
    
    
    Differences found in compared tracks.
    Zero offset detected.
    
    Comparing:
    "J:\AUDIO\Comparison\MQA\05 - Mozart A [excerpt].flac"
    "J:\AUDIO\Comparison\MQA\06 - Mozart B [excerpt].flac"
    Compared 20136440 samples.
    Differences found: 40231772 values, starting at 0:00.000000, peak: 0.0034492 at 0:21.791412, 1ch
    Channel difference peaks: 0.0034492 0.0014297
    File #1 peaks: 0.9116414 0.8876039
    File #2 peaks: 0.9112427 0.8873681
    Detected offset as 0 samples.
    
    
    
    
    
    Couldn't reliably tell the difference between the A & B versions of all 3 track excerpts (some slight differences seem to be present in track 1's A & B versions, but it appears hard for me to confidently distinguish them blind).
     
    Last edited: Jul 16, 2017
  22. missan

    missan Forum Resident

    Location:
    Stockholm
    Just listening thru the clips one will know if a blind test is meaningful, in this case I found it not to be. The differences are just too small.
     
    tmtomh, Encore and Rolltide like this.
  23. testikoff

    testikoff Forum Resident

    Re-ran my ABX test & the result was a tad better, though still unsuccessful:
    Code:
    foo_abx 2.0.2 report
    foobar2000 v1.3.15
    2017-07-16 12:42:18
    
    File A: 01 - Arnesen A [excerpt].flac
    SHA1: b0c8c1ba746d3c4b3a456522f79ceaf28800afa4
    File B: 02 - Arnesen B [excerpt].flac
    SHA1: dda477c632cbbb3053880a332fb2f046fed985cb
    
    Output:
    ASIO : Meridian USB ASIO Driver
    Crossfading: NO
    
    12:42:18 : Test started.
    12:42:59 : 01/01
    12:43:20 : 02/02
    12:43:49 : 02/03
    12:44:13 : 03/04
    12:44:35 : 04/05
    12:45:06 : 05/06
    12:45:38 : 05/07
    12:46:09 : 06/08
    12:47:02 : 07/09
    12:47:20 : 08/10
    12:47:20 : Test finished.
    
     ----------
    Total: 8/10
    Probability that you were guessing: 5.5%
    
     -- signature --
    b7032a8dfee942a9da5e549d0273bec89388c024
    
     
  24. Sevoflurane

    Sevoflurane Forum Resident

    Location:
    West Yorkshire
    These cloth ears cannot tell the difference on a first run using the Foobar ABX tool. Not sure if doing a formal test will be helpful, but I will have a go when I have time. Foobar, Oppo HA2 DAC / amp using WASAPI, Oppo PM3 headphones, West Yorkshire cochlears x2 (45 years old, the pure tone audiogram isn't pretty).
     
  25. Archimago

    Archimago Well-Known Member Thread Starter

    Wow Testikoff. Very impressive work!

    Although I have not looked at individual votes to see what the preference is, I see others have also attached their ABX scores like these. Will be very useful to see which variant of the samples is chosen - so for example in your 8/10 ABX above, I hope you've submitted your survey result with which you preferred.

    I haven't looked at the digital subtraction delta either so your post above is the first I've seen. Thanks!
     

Share This Page