MQA Listening Impressions

[PeterSt]

1 hour ago, Miska said:

I think that is better comparison point against the MQA version. At least I would myself rather listen this one than the MQA. Try it out!

 

It is not really hires, there's the sharp cut just above 20 kHz in spectrum. I made a spectrogram plot of it, but I don't post it now on your request.

 

Absolutely not. So again a nice case of not-so-hires-at-all, and completely failed with it. 6dB more compressed than the MFSL. If anything is annoying, it is this.

 

The MFSL is not really annoying to these ears and system, but with the notice that while this is from my youth and rather (should) like it, I rejected it earlier on because something seems to be not right with it. This then with the notice that I only can get it from Tidal (and DAT ). The 48 MQA version Mani posted (not 100% sure because two of that exist on Tidal) don't show this annoyance.

 

We better quit this because when we don't select the same masters in the first place, it leads to nowhere. 

 

PS: I am now listening to Percy Sledge and The Platters and that kind of worn out stuff. Troggs just coming up now (Love is all around), from Tidal. This sounds 2x better than Rush but is from the 60's. Because the tracks are so short from back then, now a "Stand by me" plays. Sounds newer than today. Really.

So Miska, I can only agree; Rush sounds annoying. But for me this not new.

[watercourse]

3 hours ago, Archimago said:

 

This is an example of muddying the waters!

 

How can you compare an MFSL mastering with typical dynamic compressed  consumer mastering and make any generalization about MQA!?

 

As Miska nicely shows us, this is like comparing a Red Delicious and a McIntosh... Both apples but different intrinsically.

 

Like I said before, given the vagueness and complexities of MQA, let's make sure we control the variables before making clajms... Either pro or against MQA.

 

I was also trying to state this, but @Archimago made the point much more clearly.

 

At this point for me, there is too much smoke and hand-waving from MQA. Even worse is what seems to be intentional misleading by use of different masters so no easy comparison can be made. I'm not going to spend any more time or money on MQA.

[Miska]

1 hour ago, PeterSt said:

Absolutely not. So again a nice case of not-so-hires-at-all, and completely failed with it. 6dB more compressed than the MFSL. If anything is annoying, it is this.

 

Yeah, but the MQA version is not the MFSL version... But the MQA version sounds like crappified version of the one I linked (and likely is).

 

Comparing MFSL and MQA versions doesn't tell anything about MQA because they are clearly from different master. The the quest is trying to find the original where the MQA version was made of.

 

[manisandher]

I thought that everyone would agree that an MFSL CD rip would be the best available redbook version of any album (derived, as it is, directly from the master tape), hence why I used it. Apparently that's not good enough to compare with an MQA derived from a presumably lesser source. OK.

 

I do agree that comparing apples with apples would be ideal, though more easily said than done.

 

However... below is a non-MQA and an MQA version of a track... from the same master. (Again captured at 24/192 from the analogue output of my DAC.)

 

Beth Orton - Moon (non-MQA):

https://drive.google.com/open?id=1KFmyilLot7dO9iPukpiJqAE2RGRcjGmX

 

Beth Orton - Moon (MQA):

https://drive.google.com/open?id=1lOQ1MUSE2Iv-kkssFwsdtNCAe1gfGKcN

 

The MQA wins... to my ears.

 

Mani.

[Dr Tone]

15 minutes ago, manisandher said:

I thought that everyone would agree that an MFSL CD rip would be the best available redbook version of any album (derived, as it is, directly from the master tape), hence why I used it. Apparently that's not good enough to compare with an MQA derived from a presumably lesser source. OK.

 

I do agree that comparing apples with apples would be ideal, though more easily said than done.

 

However... below is a non-MQA and an MQA version of a track... from the same master. (Again captured at 24/192 from the analogue output of my DAC.)

 

Beth Orton - Moon (non-MQA):

https://drive.google.com/open?id=1KFmyilLot7dO9iPukpiJqAE2RGRcjGmX

 

Beth Orton - Moon (MQA):

https://drive.google.com/open?id=1lOQ1MUSE2Iv-kkssFwsdtNCAe1gfGKcN

 

The MQA wins... to my ears.

 

Mani.

 

I have some non MFSL high res masters I prefer to MFSL masters.  You have to have the same master or you are just preferring one master over the other.

[Miska]

2 hours ago, manisandher said:

However... below is a non-MQA and an MQA version of a track... from the same master. (Again captured at 24/192 from the analogue output of my DAC.)

 

These are still different, one is 48 kHz and the other one is 44.1?

[Archimago]

2 hours ago, manisandher said:

I thought that everyone would agree that an MFSL CD rip would be the best available redbook version of any album (derived, as it is, directly from the master tape), hence why I used it. Apparently that's not good enough to compare with an MQA derived from a presumably lesser source. OK.

 

I do agree that comparing apples with apples would be ideal, though more easily said than done.

 

However... below is a non-MQA and an MQA version of a track... from the same master. (Again captured at 24/192 from the analogue output of my DAC.)

 

Beth Orton - Moon (non-MQA):

https://drive.google.com/open?id=1KFmyilLot7dO9iPukpiJqAE2RGRcjGmX

 

Beth Orton - Moon (MQA):

https://drive.google.com/open?id=1lOQ1MUSE2Iv-kkssFwsdtNCAe1gfGKcN

 

The MQA wins... to my ears.

 

Mani.

 

This is more apples to apples for sure. But as Miska noted, 44 (non-MQA) vs 48kHz (MQA) source.

 

Nonetheless, I thought the non-MQA version maybe had a "brighter" sheen to it which I thought sounded more sibilant when played side-by-side in A/B listening. Some might describe this as "more air" although I found it more fatiguing and preferred the MQA.

 

Of course we don't listen like that and the slight difference IMO is at the level where I'd be straining to hear a difference in isolation if even noticeable at all. I listened to them blind through ABX Comparator and ended up with almost exactly 50% accuracy (basically a guess) using ASUS Essence One DAC through ASIO and Sennheiser HD800 headphones.

 

Wondering what filter are you using for the non-MQA playback? Seems very weak with lots of aliasing as much as MQA! I wonder if that added to the sibilance. IMO, would be preferable if you used a standard steep filter (strong antialiasing filter) for the non-MQA one instead.

 

I see there's a 24/48 version of this album available... Try recording the output from that even though with a DR7 for this track with a relatively high noise floor, there's no reason to think it can benefit at all from >16-bit resolution.

 

 

[Miska]

I listened through same process as before, both files converted first to 44.1/32 to cut the images out.

 

The MQA version sounded again similar, more grainy and suffocated with reverb tails collapsed.

 

I use same listening approach I used for passive sonar; use your selective hearing to block out front layers like vocals and instruments and listen to the background. Just like you would try to listen what people discuss in neighboring table in a night club with loud music.

 

[bhobba]

11 hours ago, Miska said:

To me, even just 96/24 master is better than MQA version of the same. Even if reduced to 96/16 which makes it smaller/less bandwidth while providing more dynamic range than the MQA version...

 

I have read in some places its 17 bit, in others 18, but anyway they claim its way below the thermal limit - see the graph at the end of the Q&A:

  

But who knows really - they are not that forthcoming with the full detail - they make a lot of claims and give a bit away about how they do it such as explaining about the first order spline sampling (ie a triangle), whereas they actually use some other order they will not reveal the exact details of, and subtractive dither they do give the details of - but its pretty scant really.

 

So yes - if their claim is BS (ie its resolution is way below the thermal limit) then you could hear a difference with 24bit masters.   It is only above 20 bits (or dithered equivalent) I cant tell the difference - and only know of one person that can at below that.

 

So if you can hear a difference you are one of those rare people that cab hear below the noise floor or they are feeding us BS.

 

But its irrelevant anyway - was it better than 44.1/16 - which is what my claim was.   I don't know, personally, of anyone that doesn't prefer it to 44.1/16.  One person posted here he finds it all over the place - but the people I have done it with are consistent - they all prefer the first unfold of MQA to 44.1/16.

 

Thanks 

Bill

[bhobba]

10 hours ago, firedog said:

Nope, with an MQA DAC. And I don't think it is so strange. There often is very little to distinguish between even Redbook and hires. Very recording dependent.

 

Then for me that is a bit more understandable - like I said the comments I get, and what I hear personally, on both the Explorer and Direct Stream, is full unfolding is a bit thin and too polished.  I rather like it - but some do not.  The Auralic, using its own 'unfolding' sounds a bit different again - not quite as thin and polished - but some don't like it either.  Me I like both - I will need a lot more listening to see if one is better than the other - the difference didn't really jump out at me.

 

Personally I think the comparison on how MQA sounds should be with a properly done DXD master - but they are as rare as hens teeth - I have just one.  I suspect it will blow MQA away and lay bare the claims it sounds better than DXD like the 2l site does.  BTW I have tried the 2l test, and yes for me the MQA does sound a bit better than the DXD.   But I suspect it wasn't really what we call in Aus a fair dinkum comparison ie they selected stuff that MQA was better on.  Running the DXD through my 'truth' program mentioned previously shows DXD it aren't.  It was almost certianly recorded at that level - but the noise floor, frequency extension etc is not to the standard of the DXD I have.

 

Thanks

Bill

[watercourse]

1 hour ago, bhobba said:

One person posted here he finds it all over the place - but the people I have done it with are consistent - they all prefer the first unfold of MQA to 44.1/16.

 

Do these people that you are speaking for have their own way of listening and comparing MQA tracks, or only via your system, during limited-time sessions? If they have their own way, it would be interesting to hear what their thoughts are, over time, on their own systems, rather than a time- and artist-limited session.

Tidal and Audirvana+ provide a first unfold, which is what I hear in my system. Today, it was at best 50/50 how MQA tracks sounded compared to non-MQA tracks via Tidal; cumulatively, MQA falls short of sounding better half the time, IME. Most of time the MQA albums were no better than the non-MQA variant; and today, MQA and non-MQA tracks were equally preferred. I picked albums that I was familiar with, as well as new albums from Tidal's "Masters" offerings.

My iZotope filter settings are optimized for my mostly local PCM files. These filter settings sound great with non-MQA tracks, either via Tidal or locally-served. In other words, I don't think the filter settings are the problem here. I do believe that MQA's quality is inconsistent. Right now, I am actively experimenting with whether MQA has a place in my life and system (via Tidal); at this point, there's not enough of an argument in terms of consistent quality or enjoyment for me to consider it a keeper. I would rather mine Tidal's catalog than to continue buying and ripping software, but as of now, I prefer locally served files to MQA.

[PeterSt]

If we keep in mind my description of what all happens with A/D (and/or D/D) as I tried to describe in here :

https://www.computeraudiophile.com/forums/topic/37552-mqa-off-topic-spinoff/?do=findComment&comment=746622

then it is a bit beyond me how people even dare think about

1. Taking random masters instead of carefully selected ones (Mani);

2. Knotting an A/D behind it again (Mani);

3. Throw another set of filters at the again recorded result where all went wrong in the first place according to MQA (Miska);

4. Use a random DAC which processes it once again (also partly for filtering, no matter how slightly (everybody));

5. Claim that it is all a best effort (which it was), it does not sound right and thus it is not right at all (Miska).

 

4 hours ago, Archimago said:

 

Wondering what filter are you using for the non-MQA playback? Seems very weak with lots of aliasing as much as MQA!

 

Correct. And exactly the intention. So we had one thing right there. Haha.

 

4 hours ago, Archimago said:

IMO, would be preferable if you used a standard steep filter (strong antialiasing filter) for the non-MQA one instead.

 

This is dangerous territory because if "we" would have preferred that in the first place, we would have used that as a base anyway. But we don't. So it is like this or worse or a bit - up to quite a bit better, but this is the preference and that through a NOS DAC. Btw I am not Mani and it is his own choice of filter settings and more. But anyway I too like it best at this setting(s).

 

4 hours ago, Archimago said:

I wonder if that added to the sibilance.

 

I fully agree with the thesis that something is amiss there. But as I said in a previous post, it is exactly why I don't like all the Rush. Probably this whole catalog has been digitized by a means which s*cks somewhere ? The MFSL shows it the best because it is the least compressed. But the 6dB more compressed again faked Hires makes it into a larger oscillation (through these ears). Mind you, in MQA fashion.

Maybe it is no coincidence that this is (IIRC) the first MQA I see which is not Hires at all. I mean, in the base it wasn't and there is totally no improvement by means of this BUT extra headroom for metadata (FWIW). ... I now recall one other - The 88.2 Boris Blank. This too sits just in that container having explicitly not more than 44.1 in its guts. But this one sounds great to me (in MQA incarnation).

 

Especially with this Rush example, I could ask Mani to replace his Lush USB cable with the Clairixa or anything else. Try it on the MQA only and then report. After that you can try it on the MFSL as well and honestly report.

So I have to say it again : everybody who likes MQA for the better without taming hardware, likes something which kills me (OK my ears).

 

30 minutes ago I was thinking that MQA ltd works quite hard on making comparisons impossible ...

[Archimago]

7 minutes ago, PeterSt said:

...

30 minutes ago I was thinking that MQA ltd works quite hard on making comparisons impossible ...

 

Yes. That seems very much to be the case. 

[PeterSt]

11 minutes ago, watercourse said:

Do these people that you are speaking for have their own way of listening and comparing MQA tracks, or only via your system

 

This reminds me of an other maybe not coincidental matter : Bill (bhobba) is a NOS guy - and then even without any reconstruction filtering. Is that still so, Bill ?

Maybe he changed, but we really should consider how coincidental it might be that for example myself dial in only the slightest of filtering and then even in apodizing fashion. Apart from the zero ringing base filtering of course (Arc Prediction).

 

We must seriously consider how there could be two base camps - one who adheres the time domain and one who adheres the frequency domain. Not discuss either merits (please !), but how the time domain camp automatically will like MQA more easily.

[PeterSt]

22 minutes ago, watercourse said:

I do believe that MQA's quality is inconsistent.

 

First off, I think you are seriously working on this (no sarcasm !). Maybe change the A-B more into the one day this and the other day that. See how it influences your mood also knowing that my perception of matters is that it is a lot about "getting into".

Then on to the quote of your text : I believe you are right but I also believe that it is largely a matter of learning thee merits of MQA and the more you have listened to MQA the more merits you inherently know, the more you miss that in normal Redbook (remember, I never compare to Hires because it fails always anyway). What I am trying to say is : I should not have the illusion that I am improving on settings each other week. Still I like more and more of MQA. That vastly more is available now (quite suddenly) helps a lot with that. Look :

 

In the beginning I ditched Rush for what reasons ? ... we just talked about that.

In the beginning I ditched Led Zeppelin for what reasons ? I still don't know. But they all fail. Too early.

In the beginning I ditched Van der Graaf Generator. Same thing as Rush. Same era. Just doesn't work in the base. Wrongly digitized.

I did not ditch Deep Purple. It had something. Something which belongs to Rock. Sounds like a rock or something. That is MQA.

 

So for quite a long time I ditched MQA. There just weren't many good ones and there was no choice.

[Miska]

4 hours ago, bhobba said:

I have read in some places its 17 bit, in others 18, but anyway they claim its way below the thermal limit - see the graph at the end of the Q&A

 

But who knows really - they are not that forthcoming with the full detail - they make a lot of claims and give a bit away about how they do it such as explaining about the first order spline sampling (ie a triangle), whereas they actually use some other order they will not reveal the exact details of, and subtractive dither they do give the details of - but its pretty scant really.

 

It depends on the material, how much it needs bits to encode the upper band. For example on my favorite test album it varies from track to track. And with some material I'm pretty sure you can find tracks where MQA would run out of bits for encoding the top octave. I'm very curious what would happen at that point.

 

MP3 encoders tend to have similar problems with certain Metallica albums.

 

I always prefer RedBook over non-decoded MQA. With MQA unfolding it varies, but since it is proprietary process not available most of the time is sort of uninteresting niche. But I systematically always prefer the original hires over unfolded MQA, with or without MQA rendering.

 

[Miska]

2 hours ago, PeterSt said:

I did not ditch Deep Purple. It had something. Something which belongs to Rock. Sounds like a rock or something. That is MQA.

 

Compared to the original hires? I have some older RedBook master and then the new 2016 hires remaster. MQA version (if such exists) is probably made of the latter.

 

3 hours ago, PeterSt said:

4. Use a random DAC which processes it once again (also partly for filtering, no matter how slightly (everybody));

 

No, the DACs I've used don't process anything. Just conversion to analog.

 

3 hours ago, PeterSt said:

1. Taking random masters instead of carefully selected ones (Mani);

 

Just noting that none the MQA examples in this case included unfolding, so only MQA messing with word length encoding is in play.

 

And since the MQA versions used 48 kHz sampling rate, they already moved the strong images higher up a bit. To level the playing field needed to cut out the images (and by doing so improve reconstruction accuracy). Since there was no music content on the used filter transition band (that fell in the null region), doing so didn't have time domain implications in the base-band.

 

[PeterSt]

2 minutes ago, Miska said:

2 hours ago, PeterSt said:

I did not ditch Deep Purple. It had something. Something which belongs to Rock. Sounds like a rock or something. That is MQA.

 

Compared to the original hires?

 

There is no original hires of this that I know of (there is failed Hires alright). So No.

 

3 minutes ago, Miska said:

Just noting that none the MQA examples in this case included unfolding, so only MQA messing with word length encoding is in play.

 

That is your idea about it. It discards the idea of famous "deblurring".

 

[PeterSt]

18 minutes ago, Miska said:

But I systematically always prefer the original hires over unfolded MQA, with or without MQA rendering.

 

Miska, can you elaborate why you so explicitly compare with unfolded MQA ? This is not the first time you mention this.

I say that this is illegal (and also tell MQA not to say it is better already than Redbook - this is their theorie but theories s*ck some times).

 

NEVER MIND !!!

There's even word games in this. Sorry.

[Miska]

5 minutes ago, PeterSt said:

Miska, can you elaborate why you so explicitly compare with unfolded MQA ? This is not the first time you mention this.

I say that this is illegal (and also tell MQA not to say it is better already than Redbook - this is their theorie but theories s*ck some times).

 

Because they claim that it is as good as RedBook.

 

It is also illegal to compare RedBook against unfolded MQA sourced from hires. For comparing the unfolded MQA you need to use non-MQA'd version of the same hires that was used as input to the MQA encoder. Otherwise you are comparing again different masters and some algorithm that was used to convert the hires to RedBook instead of comparing MQA. Also very typically RedBook/MP3/AAC version has different mastering than hires version of the same album due to different target market.

 

So don't compare lores to mangled hires (MQA). Instead compare unmangled original hires to the mangled hires (MQA).

 

[PeterSt]

@Miska, apologies. I already tried to correct my text.

Folded, not folded, not unfolded, unfolded.

And that not in your own language.

 

[PeterSt]

3 minutes ago, Miska said:

For comparing the unfolded MQA you need to use non-MQA'd version of the same hires that was used as input to the MQA encoder.

 

Not questioning this ... But do you realize that there's another means of making MQA ... a one which does not seem to spring from Hires (just because it does not exist) and where only the expansion from 44.1 based to 48 based is used and next no mangling whatsoever took place but for this expansion. This is explicitly not the Rush example. This is (quite hard) cut at 20KHz and remains that. So those examples were hard cut at 20KHz and now are more softly cut on a little above 20KHz and now have 1950 more headroom for the filter. These are presented (unfolded) as 96.

 

I see nothing wrong with these (spectrograph) as if real Hires existed.

 

Thoughts ?

[PeterSt]

10 minutes ago, Miska said:

For comparing the unfolded MQA you need to use non-MQA'd version of the same hires that was used as input to the MQA encoder. Otherwise you are comparing again different masters and some algorithm

 

One more about this, because it is super crucial :

You maybe have found some hires you like, maybe you have bought hires you like, maybe you think that it can be bought in genuine fashion at hiresaudio.com  (we don't mention hdtracks at all) and you always have the benefit of having obtained well-done classical which genre I just don't play at all so I have no insight in that ... but :

 

What I have of that in my genres can be counted on one hand so it is non existent. The result is that to me this all (of the others) sound like crap.

See the quote again ... About all what MQA presents (whether it contains an unfold stage or not) is bearable to me. So we can put up our own theories about what is and what is not allowed, but with those theories you won't come far - just because you are right.

My point here thus is : this is all about albums which never saw Hires light. And now your theory which would be mine just the same (see quote of your text) doesn't hold; it is not applicable.

[Miska]

41 minutes ago, PeterSt said:

One more about this, because it is super crucial :

You maybe have found some hires you like, maybe you have bought hires you like, maybe you think that it can be bought in genuine fashion at hiresaudio.com  (we don't mention hdtracks at all) and you always have the benefit of having obtained well-done classical which genre I just don't play at all so I have no insight in that ... but :

 

What I have of that in my genres can be counted on one hand so it is non existent. The result is that to me this all (of the others) sound like crap.

See the quote again ... About all what MQA presents (whether it contains an unfold stage or not) is bearable to me. So we can put up our own theories about what is and what is not allowed, but with those theories you won't come far - just because you are right.

My point here thus is : this is all about albums which never saw Hires light. And now your theory which would be mine just the same (see quote of your text) doesn't hold; it is not applicable.

 

I don't think I have ever purchased classical from highresaudio.com (or HDtracks for that matter either). I have classical CD rips, SACD's and downloads from nativedsd.com. Only classical MQA stuff I have are test tracks from 2L.

 

But I do have nice hires; Mark Knopfler, David Gilmour, Roger Waters, Steven Wilson, Pink Floyd, Ritchie Blackmore's Rainbow, David Bowie, Porcupine Tree, Daft Punk, Yello, Deep Purple (the new Infinite)... Some new Steven Wilson's remixes in hires like Jethro Tull. Then bunch of various artists in Jazz, Blues and Electronic Music (EDM) genres.

 

I don't find classical interesting for MQA testing because it rarely contains notable amount of high frequency content. So not much to do for the folding. I've been buying albums that are available in both flavors, and genre-wise potentially useful for testing, first in hires and then after analysis for some of those decided to buy MQA version too because it is good test material specifically for this purpose. So my MQA buying is selective specifically to find content where I know the encoding/folding would actually have some real work to do. So finding the hard cases, not the easy ones.

 

Too bad I cannot access MQA encoder, because I have some of my own test recordings that would be good test material. Where I'm playing glockenspiel, castanets, claves, maracas, etc, close mic'ed hires. Same distance as my hand-to-ear, so I can put on headphones and compare the recording to live playing. Tracks I know and have inspected will exhibit ringing problems and such and have plenty of high-speed transients.

 

Try for example the Steven Wilson's newest To The Bone album. Hires version slowly rolls off by 30 kHz or so, looking so much like Lavry ADC that I'm guessing that's what has been used, just like his other productions too. (lyrics for the name track sort of fit the forum discussions too...  )

[manisandher]

OK, hopefully this will be the most insightful non-MQA vs. MQA comparison I've managed to come up with. The track is Christina Pluhar - Music for a While - Improvisations on Purcell - Twas within a furlong. Both the non-MQA and MQA files are clearly from the same master. As before, I'm capturing the analogue output of my DAC with a Tascam DA-3000 running at 24/192.

 

non-MQA _ HQPlayer _ poly-sinc-xtr-mp to 705.6 _ NS5:

https://drive.google.com/open?id=13p9ROlG27KuDIBpEy6X4Nw2FdXWcHFbp

 

MQA _ XXHighEnd _ MQA first unfold to 88.2 _ Arc Prediction to 705.6

https://drive.google.com/open?id=1-4ByqXm9H8c_IAYBotjsk48EHje3K4qU

 

Taking one of Peter's earlier posts into account, yes, this is very far from a fail-safe process - there's my replay chain, then my ADC, then your DAC... all having an affect on the sound. But even with all this, hopefully you'll get a sense of what I'm hearing in the MQA file. It's simply 'cleaner' sounding. And what I'm really interested in is how this 'cleaning' has been achieved with the MQA processing.

 

Mani.