MQA Deep Dive - I published music on tidal to test MQA

[danadam]

Quick question, am I the only one surprised by the following statement (in Bob's article):

FLAC is a lossless file format, a container for audio data. MQA is an advanced method for coding audio contents. MQA is normally delivered (losslessly) in a FLAC container from the music label. PCM is another type of audio that can be delivered by FLAC. Suggesting FLAC is better than MQA is like saying ‘bottles are better than wine’!

And if you decompress that FLAC you get MQA in PCM. PCM becomes the container. MQA 'turns wine into a bottle!'

If I'm not mistaken, we must distinguish between Native FLAC (which is the container) and FLAC which is the codec. And it's not for nothing that you can put FLAC in an Ogg container https://xiph.org/flac/faq.html

 

[Zensō]

Today's only model that seems to work is distributing music in the interest of selling something else. Apple as I mentioned, sells music so that it can sell far more profitable hardware. Amazon does it so that you buy millions of other items they sell and keeping you a Prime customer.

This ^. In an interview at the Code conference, Walt Mossberg asked Jeff Bezos why Amazon moved into streaming video. He said they did it “to sell more tennis shoes”, then went on to explain that their add-on services were there to keep people in the Amazon Prime ecosystem. It appears music streaming serves the same function for Amazon, and possibly to an even larger degree for Apple (Apple Music is now included in the Apple One services bundle along with Apple TV+, Apple Arcade, Apple Fitness, iCloud storage, etc.)

Some small number of niche streaming services may survive, but going forward, the major players are very likely to be the big tech firms using music as a loss leader. I believe this is likely to be a death sentence for Tidal/MQA, and possibly even Spotify.

 

[Thomas savage]

This isn't a MQA deep dive , its a bounce dive at best .

Diving related trolling , you can't touch this ..

 

[Grooved]

A reduction factor of two is irrelevant. 10 is the goal (and the reason why MP3 was so successful initially).

And it's only for above 48kHz MQA files, all 16bit MQA and 24bit MQA 44.1 or 48kHz have almost the same file size than FLAC
It's from 24/88.2 that you start to get smaller size with MQA

The idea is how do we prove MQA is better than CD quality at the output of a DAC?

If you want to compare MQA with CD, it has to be done with 16bit MQA file then. It would not be fair to compare CD and 24bit MQA

Would be interesting to see what the spectrum of the MQA core decoded version looks like

Tested on the 2nd track of Terry Riley 02 - Kronos Quartet - Sun Rings- Hero Danger that @jensgk bought, and that appears to be a fake Hi-Res

Same thing on Qobuz and Tidal
24/96 FLAC from Qobuz :

24/48 MQA from Tidal decoded to 24/96 :

 

[mSpot]

I think MQA has made a strategic mistake of never engaging the audiophile community on these forums.

You think it's a mistake? I think it is a very intentional and calculated strategy.

Instead they only went the route of one-shot interviews with press and such with no allowance from the other side to ask tough questions.

This is completely consistent with not engaging on forums. They want to control the narrative and avoid tough questions.

 

[Grooved]

...
Son of gun! The actual (live) recording is at 44.1 kHz but is somehow mixed into a high-res envelop. We know this because some noise has been added there including that one peak at nearly 48 kHz.

The most shocking part is who mastered this: the famous Bob Ludwig!!! From the credits: "Mastered by Robert C. Ludwig "
...

Amir, I'm the first to agree with you on the subject of avoiding generalizing (like I said above to specify 16bit MQA and 24bit MQA) and to not turn something said in a post here into a truth without facts, but in this sense, I think you should edit your previous post to clearly write that it's Bob Ludwig that mastered this album, but we can't know who created this conversion of the master.
The goal is not to defend Bob Ludwig as it's possible he was the one who created this conversion, but fact for now is that we don't know who created this file. And he's not responsible for any mess done during mixing if that's the case.
I can't be sure too that's not him that created it, but what makes me having doubts is, for one example, that he co-worked on Daft Punk "Random Access Memories" and that this album is certainly one the recent Grammy awards that truly deserved it regarding what Grammy said about what a recording should be, and it's only my point of view, but strangely, some tracks of this album are ones that I find the easiest tracks to do an ABX test and find difference between AAC and 16/44.1 FLAC.

I need to repeat that for most artists, there's a choice for the label regarding the recording studio, the recording engineers, the mastering studio, the mastering engineer, who is the one who creates the original master, and that then, the master is propriety of the label... and that label can call later a different mastering studio, which I would not always call a mastering studio but a creator of files, to prepared it or in this case create a conversion and a fake HiRes file to give to streaming providers.

I hope you can understand that

 

[mSpot]

Tested on the 2nd track of Terry Riley 02 - Kronos Quartet - Sun Rings- Hero Danger that @jensgk bought, and that appears to be a fake Hi-Res

However, see Amir's comment:

Figured out what is going on. The filtering depends on what segment of the track you are playing! Clearly some of the instruments are recorded with limited bandwidth, and others not. What a mess....

 

[Grooved]

However, see Amir's comment:

​

If you have quoted the line just under that, it would look more like I was adding the fact that it's the same problem on Qobuz and Tidal than on both websites they used to get it, but I should have been clearer ;-)

 

[levimax]

Looks like that nails it. In this day and age, you have to meat your customers at eye level. We're not the dumb sheep we used to be in the last century, notably pre-internet.

It would be interesting to see if there is any scientific backup for this statement. From my perspective, including living during a decent part of the last century, I would conclude the opposite. To me it appears the internet has dumbed down most people when it comes to technical issues and they more or less go along with what ever the "influencers" are peddling much of which is pseudoscience at best. Back in the early Hi-Fi days of the 1950's and 60's people were really engaged in Hi-Fi as a hobby and read up and studied and often built their own equipment.

 

[stopbeingalentil]

From a purely coding point of view, yes, the extra efficiency is not significant. MQA though includes other bits such as a different reconstruction filter for the DAC that they hang their hat on as being better. I have not discussed, nor care about that feature if you can call it a feature. And at any rate, if you software decode MQA, you don't get that anyway (that is how I use it).

They also say some stuff about optimizing the file post ADC which relies on a pure new digitization of audio using this scheme which I have not seen (rather, know about) used to date in any MQA content.

So the total feature list if you call it that, goes beyond the main focus of OP and what we are talking about in this thread.

Except GoldenOne addressed both of those points in his videos, namely trying to find the closest known match for the hardware reconstruction filter, which he did in his first video, and pointing out that in a modern music production there's likely no single but many ADCs involved - or none at all in the case of purely electronic music, hence that whole "compensating for imperfections introduced by the ADC used" claim would - if at all - only apply to an end-to-end MQA production and cannot apply to the vast majority of the MQA catalog that just got converted in a generalized batch processing

I don't get why you're so eager to regurgitate their marketing speak, all for which they provide zero proof but instead ask for appeal to authority in the form of bob stuarts credentials, when at the same time you also keep insisting for everyone else in this thread to stick to facts whenever someone offers an oppinion - even if they clearly qualify it as such

 

[amirm]

Except GoldenOne addressed both of those points in his videos, namely trying to find the closest known match for the hardware reconstruction filter, which he did in his first video, and pointing out that in a modern music production there's likely no single but many ADCs involved - or none at all in the case of purely electronic music, hence that whole "compensating for imperfections introduced by the ADC used" claim would - if at all - only apply to an end-to-end MQA production and cannot apply to the vast majority of the MQA catalog that just got converted in a generalized batch processing

He had no ability to prove this point and simply repeated one of the talking points against this MQA feature. That would require investigative work across a ton of MQA content to know what is or is not there. His experience of using an automated system which is reserved for smaller distributors is not indicative of the larger usage of MQA that involves MQA and a few major mastering house.

Given all of this, OP's main work was to encode test tones and show that the encoder could not handle them. And that is what we are examining here, not him repeating an argument he has heard. Remember what I said:

"So the total feature list if you call it that, goes beyond the main focus of OP and what we are talking about in this thread. "

 

[Vladimir Filevski]

from the 'papers' i have read on AES, they're borderline a predatory journal.
the papers' transcripts follow no real academic structure, their language is intentionally difficult to understand, and their abstracts have no real meaningful information and is there to look 'sciency'
At least the name of the journal represents that, they're a bunch of engineers who wanted to do 'science'.

Total and utter nonsense.

 

[amirm]

I don't get why you're so eager to regurgitate their marketing speak, all for which they provide zero proof but instead ask for appeal to authority in the form of bob stuarts credentials, when at the same time you also keep insisting for everyone else in this thread to stick to facts whenever someone offers an oppinion - even if they clearly qualify it as such

It is this kind of information-free accusatory tone that drags these threads to closure. I have provided a ton of technical information, objective analysis and explanation of both the features and market dynamics of MQA. There was no "appeal to authority." If you think that, you have let your illogical emotional outburst bloody your eyes so much that you can't read what is posted.

This has been the first time were we have managed to somewhat stay on the objective analysis of what is in front of us. But a few of you can't resist posting such information-free nonsense. Do it again and I will issue a Reply ban to you.

 

[amirm]

from the 'papers' i have read on AES, they're borderline a predatory journal.

How many is that?

Have you for example read this paper:

Or:

 

[Hai-Fri. Audio]

This has been the first time were we have managed to somewhat stay on the objective analysis of what is in front of us. But a few of you can't resist posting such information-free nonsense.

1 step forward...

 

[Raindog123]

Something you didn't mention but I value is elegance in efficient coding of music. I have always considered PCM format to be highly wasteful. As a person who has spent decades optimizing technology, it seems like such a poor solution. Going from 44.1 kHz to 88.2 kHz doubles the data rate yet there is hardly any musical information to be gained from that doubling. In that regard, MQA's approach of noticing the statistical aspects of music and encoding that is appealing to me. It is simply neat!

1) Modern ‘general purpose’ compressors are extremely efficient with low-information-density cyclostationary data. An 88kHz FLAC carrying 44kHz ‘plus one bit’ of PCM-modulated waveform satisfies that nicely. Resulting in the 24/88 music FLACs being comparable in size to 24/44 or 24/48 MQAs. Which was pointed at and illustrated by actual file sizes a few times within this thread.

2) ‘Noticing the statistical aspects of music‘ while ignoring ‘hearing aspects of the listener’ - ie not being able to hear ultrasonics, and especially within 6dB above noise, is hardly ‘neat‘ in my playbook. [If i remember, you pointed this yourself some posts earlier.]

 

[DimitryZ]

New member here, involved in a long running parallel thread on pinkfishmedia.

I think this simple look can elucidate the great lossy/lossless debate. We are leaving MQA's "deblurring" process out for now, since this a separate debate/controversy.

Let's put some numbers on the distortion expected in the types of codecs helpfully explained by Amir. In engineering, a useful excercise is called "order of magnitude analysis." It allows one to get approximate understanding of system behavior.

Mathematically Lossless Codec:

Mathematically perfect algorithm running on a real-world computer and network, will still have an error once in a great while. Let's pick a really tiny value - 1 in a billion or 1E-9. This results in the error against the original file of -180dB. Great performance!

Lossy codec:

Let's stipulate a very good algorithm that delivers an error of 1/10 of 1% or 0.001. This results in the error against the original file -60dB. Not bad at all!

MQA:

From graphs posted at PFM and elsewhere, "eyeball" MQA null against LPCM looks like -120dB, or 1 in a million (1E-6). This is clearly better than an excellent lossy codec but clearly not as good as mathematically lossless one.
https://drive.google.com/file/d/1o79kijUug1Rg3Ne0aHM35oibMMPbWRK6/view?usp=drivesdk
(Red and Blue curves used for conservatively deriving 120dB number. Yellow vs. Green maybe more relevant and seem to overlap up to 35 KHz even more. If properly averaged and weighted the final number is likely to be smaller - i.e better null).


Let's consider what above numbers mean in the context of consumer audio replay:

Leaving aside a question of when do deviations become audible, one still needs a sound system that has amplification with low enough distortion and noise to actually playback these differences.

I own modern amplifiers with very low distortion and very high power - Emotiva XPA-1 gen.2. Their specifications list Signal to Noise at 89 dB at 1 watt and 117dB at full power (600w). Let's take an average of 103dB. Added distortion, though very small (-80dB), will make the total number worse.

So on my very low distortion system, I could at least theoretically be able to hear distortion in the lossy codec. However, both mathematically lossless and MQA distortion will be below the system's noise/distortion floor and, therefore, inaudible.

They will be entirely indistinguishable from one another to the listener and are, therefore, identical. The big debate about MQA lossy/lossless status is entirely meaningless for consumer reproduction.

I note that a lot of MQA criticism is essentially someone showing you some hard to see plots and insinuating that *something* is bad. In engineering we call it "arm waving." Once you put some numbers on the observed phenomena and place it in the context of actual system operational usage, one can make informed conclusions. Simple excercise above shows that mathematically lossless and MQA are identical in the context of home listening.

I welcome mansr and others to offer a competing simple analysis that challenges above numbers in a meaningful and substantial way.

 

[dc655321]

From graphs posted at PFM. and elsewhere, "eyeball" MQA null against LPCM looks like -120dB, or 1 in a million (1E-6). This is clearly better than an excellent lossy codec but clearly not as good as mathematically lossless one.

Links to this data and analysis?

In engineering we call it "arm waving."

You're doing your own fair share of "arm waving" with this hollow post...

 

[DimitryZ]

Links to this data and analysis?



You're doing your own fair share of "arm waving" with this hollow post...

No welcome?

Link to the plot of MQA/LPCM comparison added to above. I assumed you folks were up on this stuff.

Other stipulations are in my post. Analysis is simple dB math for amplitudes, which I assume all can do:

dB=20*log(A1/A2)

"Hollow post?" It's one of very few with actual numbers - audioSCIENCEreview?

 

[Music1969]

I think MQA has made a strategic mistake of never engaging the audiophile community on these forums. Had they done so, they could have provided the explanations I provided in this thread, a long time ago, obviating the need to do the video OP did, and tons of arguments back and forth. Instead they only went the route of one-shot interviews with press and such with no allowance from the other side to ask tough questions.

Since Amazon and Apple and Spotify are now using lossless/hi-res to differentiate , I think if Tidal are really 'all in' with MQA as the main MQA streaming services (there's small amount of MQA on Qobuz by 2L) they need to take on some responsibility and take on some of the messaging work.

Either share the marketing responsibility with MQA Ltd or dump MQA completely or just they (Tidal) can just watch their subscriber numbers disappear.

Tidal can't be half assed with the big boys coming for their necks. Especially with the huge recent game changer with pricing - Amazon Music HD and Apple Lossless will be same price as Spotify's current lossy service. The pressure is really on for Tidal's survival and hence MQA Ltd's survival.

We still don't know what move Google will make with YouTube Music.

Tidal actually just recently re-jigged their pricing, having the MQA tier priced higher than CD quality tier . But that will need to be re-done with the recent Amazon Music HD and Apple Hi-res Lossless pricing news.

I personally don't want MQA anywhere but will be interesting to watch what Tidal do with both pricing and MQA.

Strap yourselves in for a wild ride ahead