Thoughts on a Method to Compare MQA and Non-MQA Files

[Jud]

Some of the comments in the enormous MQA "vaporware" thread, including some of mine, discuss listening comparisons of MQA versus hi res or Redbook.  Some recent listening tests I've done, together with my reading in the academic literature, indicate to me that memory, even short term memory, doesn't do the best job of accurately preserving the details of the way things sound.  I wanted a way to listen to two tracks *simultaneously*, avoiding any loss in accuracy due to memory.

 

@pkane2001 has done a wonderful job of building a tool that allows such simultaneous comparisons (and much more besides), called DeltaWave Comparator ( https://deltaw.org/ ).

 

The method I'll be suggesting uses two of DeltaWave's many capabilities.

 

- DeltaWave allows you to select two tracks for simultaneous comparison, mixing each stereo track down to mono, then assigning one track to the left channel and one to the right channel.  I believe (though I haven't had a chance to test yet) that the newest version of the software allows the reference to be either the left or the right channel, a key point.

 

- DeltaWave also allows you to save this left-right comparison as a WAV file.

 

The general comparison method is to select an MQA track and non-MQA equivalent, then save two WAV files from the DeltaWave left-right comparison, one with the MQA track on the left, one with it on the right.  I'm going to suggest some ideas regarding playing these files back with Audirvana+ or HQPlayer.

 

Now what about matching file resolutions, since MQA first shows up as lower resolution, then is "unfolded" and treated as higher resolution by software and DAC?  I would suggest two different comparisons here, particularly since MQA claims its processing won't harm the sound with non-MQA-enabled DACs/software: First comparison, treat the MQA file as an ordinary file of Redbook or whatever other lower resolution it's packaged as, which is very easy - the process of mixing down the stereo file to mono in DeltaWave will accomplish this on its own.  Second comparison, decode the MQA file to its "full resolution" (using the Bluesound decoder if you have it - any other suggested methods, @mansr or @Miska?), save as a digital file using one of the available software virtual sound cards, then run it through the DeltaWave mixdown-comparison-save process.

 

We've now got a matrix of 4 comparisons to save as WAV files  - left reference and right reference for both original MQA resolution and unfolded resolution.

 

We're going to add one more variable, which will not increase the number of saved files, but will increase the total number of comparisons to 8.  Each of the saved files should be played back in two ways with A+ or HQP: (1) using no upsampling with MQA-enabled DACs/software, or upsampling filters as close as possible to MQA's for non-MQA-enabled DACs/software, and (2) using upsampling filters that will minimize any distortion from content in the low ultrasonic range.  @mansr, can you suggest A+ Sox filter settings for each of these, and @Miska, can you suggest which HQP filters might be chosen for each?

 

Thus for each track we'd have the following comparisons: left and right for each of the original and "unfolded" MQA resolutions, played back with MQA-type and normal filtering.

 

To blind the comparison, choose random order playback of the 4 files (which I know is available in A+, but don't recall regarding HQP).  If your preference for left or right turns out to have switched depending on whether that channel was playing the MQA track, then you like MQA or non-MQA better, depending on your choice.  (Edit: Or perhaps the mastering is different, and you like one mastering better than the other.)  If your choice remains in the same channel, then whatever preference you might have between MQA and non-MQA was less than the difference between your left and right ears, or your left and right speakers as positioned in your listening room.  If your choice switches, but at random, then you're not hearing enough of a distinction between MQA and non-MQA to make an audible difference to you.

 

A couple of notes regarding use of DeltaWave for these comparisons:

 

- If you want to avoid your choices being biased by what you see in DeltaWave's analysis of the files before playback, then turn off the various ways to view the file comparisons in DeltaWave's settings.

 

- Since it is likely the MQA file will have used minimum phase filters during ADC, and the regular versions may have been done with linear phase filters, I would suggest turning off phase drift correction in DeltaWave.

 

This is a fair amount to go through for each comparison of an MQA and non-MQA track, and not everyone may have the means to decode MQA files.  For this reason, I'd like to request that those who do have such means make 30-second snippets of non-MQA and MQA tracks (non-decoded and decoded) available for the purpose of testing by others on request.  One set of files I'd be interested in having such samples of if available, since they've been discussed by a few people in the "vaporware" thread, would be a track or two of Peter McGrath's.

 

That's (finally) it, everyone!  Feel free to discuss, dispute, or otherwise here or in the "vaporware" thread.

[Jud]

I don't have Roon, so can't say.  I'm sure others may have suggestions.

[Jud]

17 minutes ago, Lee Scoggins said:

Aren't you inherently losing the imaging information by reducing to a single channel?

 

That would seem to be a major flaw in the approach.

 

Some, but plenty remains. There's soundstage height and depth, coherence, size, and vertical location of (one half of) each instrument and/or vocalist....  In fact I find such comparisons often turn on these very characteristics.

[Jud]

30 minutes ago, PeterSt said:

 

 

😁

[Jud]

1 hour ago, sandyk said:

 

 Sorry Jud, but with this one I have to agree with Lee.

We need 3 versions as you mentioned , the original high res file, the non Decoded MQA, and the decoded MQA version. These should be made available via the PM system to prevent abuse or excessive demands by other than interested forum members , and the listening results tabulated.

 

Hi Alex - If you have a look at what I laid out, I actually mention not only those 3, but a 4th, a Redbook version to match the resolution of the non-decoded MQA.  Whether people want to tabulate results is up to them.

 

Quote

Soundstage HEIGHT from a MQA recording ?

 Very few recordings exhibit a good illusion of height, and those that do are highly unlikely to be available from MQA. (" The Storm" from a Hybrid Chesky SACD is a good example, and even then your gear needs to be way above average to demonstrate this properly .) 

 

My gear must be way above average.  Must be the USB DAC!   

 

Edit: For example, listen to the Steve Hoffman-mastered DCC gold CD of Pet Sounds.  Tremendous soundstage height from a mono recording.

[Jud]

3 hours ago, sandyk said:

 Please explain how it is possible to get HEIGHT information from a Mono recording.

 

The same way you do from a stereo recording without having another pair of speakers at ceiling level - as you mentioned, phase (due to recording mic placement or added effects); and also speaker-room interactions.

 

But speaker-room interactions should give similar soundstage height on virtually all recordings, and that's not the case (for example, on Tom Waits' very industrial version of the Disney chestnut Heigh Ho, it's quite squashed) so I think a fair amount of it at least on my system is due to recording mic placement or effects.

[Jud]

4 hours ago, sandyk said:

The record companies need to see that many people have heard and don't like what MQA is doing to their favourite material ,and that their reported findings have been widely read by not only members of large Audio forums, but seen by many other casual readers of these forums. Your methodology will go over the head of most members of the general public, and I doubt that it will be as revealing as listening directly via a highly revealing system.

 

It *is* listening directly to both, rather than listening to one while trying to remember another.

 

I agree with what you say about the procedure being more fuss than the vast majority of people want to go to.  But I wasn't after "easy," I was after "What will eliminate most variables from the comparison other than the absolute sound quality of the MQA versions (encoded and decoded) versus the mathematically lossless versions (Redbook and hi res)?"  And I agree fewer people will mean less of an impression on the MQA folks, and though I regret that, I can't help it, since as I say, I wasn't formulating this process to appeal to the greatest number of people.

 

I am hoping community-minded folks will make reasonable length (say 30 seconds) samples of tracks available for testing in the various resolutions needed, to eliminate some of the fuss and bother for the rest.

[Jud]

2 minutes ago, sandyk said:

Jud

 I am not sure that we are on the same page.

 A pure MONO recording as with very early recordings was made using a single microphone and is  played back these days through 2 front speakers, so there should be identical information going into both speakers with no height ,depth or width information, just a central image .

 

That might well be the case, but I know it wasn't with the Beatles mono recordings, and I'm fairly certain it wasn't with Pet Sounds.  I also wonder whether it was the case (my guess is no) on the Phil Spector mono recordings I have.

[Jud]

7 hours ago, Lee Scoggins said:

 

I believe there is an AES standard or research that suggests a certain number of seconds is good for aural.memory.  Maybe you could break tracks into short cuts that are played sequentially...?

 

Already did.  Though informal, a statistical analysis showed people lost detailed information about things like the harmonics that distinguish two acoustic guitars from each other under such a scenario with statistical significance level of p=0.06.  So you create a dilemma: Short enough to remember more accurately, too short to pick up the fine details that people believe distinguish good systems from each other.  There is also academic research that shows in the ABX situation, the "B" sample replaces the "A" sample in the brain's echoic memory store, so you can't even accurately recall what notes were played in the "A" sample, let alone fine details.  See for example http://deutsch.ucsd.edu/psychology/pages.php?i=209 .

 

So go ahead and tell me all about the possibility of a bit of missing soundstage information when using simultaneous playback. I'll take it over the vagaries of (lack of) acoustic memory any day.

[Jud]

Just now, pkane2001 said:

 

Gents, I like the idea of doing your own testing and not relying on someone else to tell you that it sounds better. That's one of the reasons I developed DeltaWave. The comparator in DW is one tool where I intend to keep adding functionality and new ideas to improve on audio evaluation techniques. The tool supports the two-channel simultaneous comparison that Jud is proposing here, as well as standard ABX test, and a subjective preference test.

 

What's more the tool allows any of the tests to be run in 'learning' mode where you can see and tell what track is playing, and to then repeat the test in blind mode, where the tracks are randomized. DW also creates a signed report of your blind test, including hashes of the files and all processing DW itself did to match them, so these can be validated by others.

 

You can do a sequenced A/B test, A/B/X test, a simultaneous Stereo X-Y test, or a sequential preference test and get the resulting statistics that demonstrate that your result was not produced by guessing or random choice.

 

DeltaWave is still under development and probably will be for a long time (I like to tinker) but basic functions are there for anyone to try.

 

Regards,

     -Paul

 

In other words, the software allows you to use both methods and compare results.   

 

So which do you perform better at, @Lee Scoggins - the test you've suggested or the one I suggest?

[Jud]

1 hour ago, Miska said:

 

It would be interesting to know how well MQA supports multichannel configurations. 2.1, 3.0, 3.1, quad, 5.0, 5.1... Such usually need channel mapping at minimum.

 

 

It doesn't seem to support 2.0 that well yet.  😉

[Jud]

7 hours ago, Paul R said:

 

Both of you guys are probably speaking true, but I admit, I don't know how you can have depth or height with just a mono recording. Would not any such artifacts be only from the speaker or speakers? 

 

I do not listen to very many monoral recordings, and when I do they sound clear and center placed to me, no real soundstage such as with stereo, just more like a phantom center channel. 

 

-Paul 

 

 

Here's a question: How is information about vertical position embedded in differences between left and right channels?

 

By the way, thought of another way to create an impression of height on a mono recording: Reverberation, natural or as an added effect, creates the impression of an overall larger (therefore higher) space.  With that in mind, listen to the famous Coke can "percussion" on "God Only Knows" from Pet Sounds.

[Jud]

37 minutes ago, Jud said:

By the way, thought of another way to create an impression of height on a mono recording: Reverberation, natural or as an added effect, creates the impression of an overall larger (therefore higher) space.  With that in mind, listen to the famous Coke can "percussion" on "God Only Knows" from Pet Sounds.

 

In addition, see the following: https://theproaudiofiles.com/width-height-depth-in-a-mix/

 

This explains why we might tend to perceive vocals, particularly higher range vocals, as located at a greater height than a bass drum or kick drum.

 

Note that neither this nor reverberation requires different information from left and right channels, i.e., mono will serve just as well as stereo to give these sorts of height cues.

[Jud]

1 hour ago, sandyk said:

 

Don't forget too, that most speakers have the tweeter(s) and LF driver at different heights on the baffle.

 

I take it you did not read the linked article, since that's exactly what it says.   

[Jud]

3 hours ago, Lee Scoggins said:

t seems like two channels would facilitate an understanding of what the deblurring filters do.

 

Speaking of admiring creativity:

 

People objected that the dual-mono test would affect soundstage height. Several mechanisms have now been discussed, and not a single one requires two different channels.

 

Now you propose that 2 speakers with differing material are needed to detect filter ringing, and intermodulation and harmonic distortion.  What possible mechanism would you suggest for this?

 

I'm very happy to hear cogent criticism of the proposed method, or have people try it out versus sequential listening and see where they have more success.  But rank speculation just wastes time.  So please provide a reasonable mechanism (you or anyone else), preferably with some support, or let's drop it.