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

[krabapple]

i didn't mean to be scornful, but it is what it is.

The science engineers do is useless if they can't express it properly. if you want to partake in academia and publish papers then you need to be part of academia and receive proper training to do so.

the scientific community is RUTHLESS when it comes to writing, because if you fail to communicate the results of a 100k$ experiment then that's 100k$ wasted, if you leave 'space' for interpenetration then your entire research could be invalidated.

LOL. Can you replicate an experiment from a Nature methods section? You'll have to dig well into the 'supplemental materials' to even try. And even then, good luck!

Your generalizations strike me as crude. But I'll join in to the extent that IME, engineers online are entirely too convinced that they stand at some Olympian vantage on The Truth about nature. I shudder when they e.g., venture to 'debate' evolutionary biology.

 

[tmtomh]

See my explanation with one of my own 2Fs recordings how audio information with frequencies higher than half the the 1Fs sample rate can be packed into the low bits of a 24-bit file sampled at 1Fs without compromising the recording's analog noise floor:
https://www.audiosciencereview.com/...-music-on-tidal-to-test-mqa.22549/post-759747
and
https://www.audiosciencereview.com/...-music-on-tidal-to-test-mqa.22549/post-760938

This is a technique called "steganography" - see https://www.researchgate.net/publication/45949372_Steganography-The_Art_of_Hiding_Data

John Atkinson
Technical Editor, Stereophile

I - and I'm sure many others - would be interested to hear your thoughts on why preserving the ultrasonics is a desirable thing to do. In other words, in your view, what is the advantage to storing some portion of the 22.05k-44.1k (or 24k-48k) frequency information in the lower bits of the file, versus just not retaining those >22.05k/>24k frequencies at all?

This is not a sarcastic or "gotcha" question. While my views on this question are well-known in this thread, I'd genuinely like to know yours, and I'm sure I'm not alone.

Thank you.

 

[krabapple]

Most definitely. The most authoritative paper, and analysis actually comes from Bob Stuart! It is both a conference paper and Journal with the latter being more detailed. It is also available in public. It is called Coding for High Resolution Audio Systems

https://www.aes.org/e-lib/browse.cfm?elib=12986

I found an online version although I don't know if it is the conference paper or the journal one: http://decoy.iki.fi/dsound/ambisonic/motherlode/source/coding2.pdf

"What do we mean by high resolution? The recording and replay chain is reviewed from the viewpoints of digital audio engineering and human psychoacoustics. An attempt is made to define high resolution and to identify the characteristics of a transparent digital audio channel. The theory and practice of selecting high sample rates such as 96 kHz and word lengths of up to 24 bit are examined. The relative importance of sampling rate and word size at various points in the recording, mastering, transmission, and replay chain is discussed. Encoding methods that can achieve high resolution are examined and compared, and the advantages of schemes such as lossless coding, noise shaping, oversampling, and matched preemphasis with noise shaping are described. "


This paper is heavily referenced in other research. And if you read it, it actually has a very sober look of the audio saying we don't need crazy sample rates to preserve what is important in audio. Likely the reason MQA doesn't even try to encode above 88/96 kHz and just upsamples.

The referenced research re: audibility of >22kHz frequencies was thin at best.. Oohashi et al? Please.

The 'need' for hi rez at consumer playback was a struggle to establish. This is what Lipshitz, Vanderkooy et al objected to in a later issue of JAES.

 

[DimitryZ]

it's like we're reinventing the history of lossy audio codecs here....

mp3 was released in 1993!

I'm glad people are finally realizing how lossy audio codecs were designed (i.e. , using psymodels of human hearing...with tradeoffs between file size and perceptual transparency) . But disturbed that it's news to so many. It does explain the irrational and ignorant dismissal of them among so many 'audiophiles' though.

I shouldn't admit it here, but I listen to satellite radio, which surely doesn't have a respectable bit rate. But I like Marcus Miller and his weekly show on Real Jazz channel.

But then again I listen to minidiscs and...8-track tapes.

 

[lucretius]

I see 70 million total, so only 2% MQA. Doesn't seem to be a concern to me, but if it does to you, there are now three non-MQA subscription services.

Lot's of good choices!

A few years back, I remember reading Warner was planning to convert its entire catalog to MQA. There where similar statements from Universal and Sony but these two seem to have pulled back. Also, it seems Warner has slowed down a bit -- I'm not sure what can be inferred by that. Perhaps non-MQA streaming is safe (as in it will continue to exist) -- but it did not seem like that back then.

 

[mSpot]

Tidal says they have 70 mullion tracks. If that 1.5 M is correct, then 2% of content on Tidal is MQA which is negligible. Hence my experience of hardly ever seeing MQA content unless I seek it out.

The overwhelming majority of MQA tracks are from Warner Music. The reason is that Mike Jbara, the CEO of MQA is from Warner Music. I wonder what the count would be without Warner. Note that in the MQA album list, many of the labels are owned by Warner. https://en.wikipedia.org/wiki/List_of_Warner_Music_Group_labels

 

[DimitryZ]

The referenced research re: audibility of >22kHz frequencies was thin at best.. Oohashi et al? Please.

The 'need' for hi rez at consumer playback was a struggle to establish. This is what Lipshitz, Vanderkooy et al objected to in a later issue of JAES.

This seems like a nice overview that references your references. Enjoying it now.

http://www.aes.org/e-lib/browse.cfm?elib=20455

 

[tmtomh]

The referenced research re: audibility of >22kHz frequencies was thin at best.. Oohashi et al? Please.

The 'need' for hi rez at consumer playback was a struggle to establish. This is what Lipshitz, Vanderkooy et al objected to in a later issue of JAES.

There's actually some very sensible content in Stuart's AES paper, some of which his acolytes and other audiophile influencers would do well to heed: for example he says 20 bits of depth is sufficient and he makes no argument for any sample rate higher than 52kHz.

But yes, you are correct, there's a lot of nonsense there too, especially for a peer-reviewed paper (although with the caveat @amirm mentioned that we don't know if the freely available version is identical to the peer-reviewed version or if it's different in some way). In addition to relying on the discredited, never-replicated Oohashi study (as everyone who claims ultrasonics are heard or "sensed" must, since it's the only source there is), his paper is also filled with turns of phrase like "it has been suggested" and "the author has been in listening tests where" - unscientific, anecdotal, statements that form the basis for his claims about particular types of reconstruction filters and the supposed inadequacy of a 22.05k Nyquist limit. I'm not an expert in the field, but I am an academic, and it's kind of hard to believe this sort of thing would pass muster in a peer-review process.

 

[lucretius]

here's actually some very sensible content in Stuart's AES paper, some of which his acolytes and other audiophile influencers would do well to heed: for example he says 20 bits of depth is sufficient and he makes no argument for any sample rate higher than 52kHz.

This is one of the few things I can agree with. I am curious about the 52 kHz vs say 48 kHz.

 

[awdeeoh]

A few years back, I remember reading Warner was planning to convert its entire catalog to MQA. There where similar statements from Universal and Sony but these two seem to have pulled back. Also, it seems Warner has slowed down a bit -- I'm not sure what can be inferred by that. Perhaps non-MQA streaming is safe (as in it will continue to exist) -- but it did not seem like that back then.

Who knows, those 16/44.1 are deblurred by MQA in plain PCM. No MQA metadata.

It does not require MQA DAC or Software Decoder and people are loving it cause there's no MQA light in it and then they spout how MQA sound quality is worse. lol

 

[Newman]

The referenced research re: audibility of >22kHz frequencies was thin at best.. Oohashi et al? Please.

Oohashi has said that he has no evidence of hearing above 20kHz. He attributes the Hypersonic Effect to other, unexplained, causes.

 

[amirm]

The overwhelming majority of MQA tracks are from Warner Music. The reason is that Mike Jbara, the CEO of MQA is from Warner Music. I wonder what the count would be without Warner. Note that in the MQA album list, many of the labels are owned by Warner. https://en.wikipedia.org/wiki/List_of_Warner_Music_Group_labels

In general, when it comes to online delivery of music and movies, the major labels and studios don't do any encoding themselves. The contract with labels to get their content is the just right to distribute. You are on your own to get the bits.

The distributor can do the encoding or go to service providers that have the content encoded and available in different formats. I would be very surprised to see Warner doing any encoding of MQA themselves. Likely this work is being done by MQA itself.

 

[oursmagenta]

Your phone would not work without this technique! Spread spectrum is used to encode a signal among a ton of RF noise and then recovered. Correlated data is easy to distinguish from uncorrelated noise. Add some redundancy and you can recover pathological cases. What seems random to the eye is not random to cross-correlation algorithms.

So you are saying that the stuffing process is additive?
Like Final_PCM{lowest bits} = Original_PCM{lowest bits} + MQA_packet and not Final_PCM{lowest bits} = MQA_packet?

 

[amirm]

So you are saying that the stuffing process is additive?
Like Final_PCM{lowest bits} = Original_PCM{lowest bits} + MQA_packet and not Final_PCM{lowest bits} = MQA_packet?

No, I am just commenting on "difficulty" of embedding data in random bits. I have not looked to see what MQA does.

 

[DimitryZ]

And let's not forget the contribution to a version of spread spectrum by the beautiful Hedy Lamar in 1941

https://www.aps.org/publications/apsnews/201106/physicshistory.cfm

 

[ebslo]

the supposed inadequacy of a 22.05k Nyquist limit.

His plots showing the limit of audibility of 20kHz content at around 90dB SPL (-30dBFS where FS is 120dB SPL) are interesting when compared to the "corpus music" plot that claims "music" doesn't contain any content higher than -30dBFS at 20kHz. It seem by his research, "music" contains no audible content even at 20kHz.

 

[abdo123]

LOL. Can you replicate an experiment from a Nature methods section? You'll have to dig well into the 'supplemental materials' to even try. And even then, good luck

If i study the same topic I should, that’s the point. Unlike the abstract, the materials and methods + supplementary materials are not meant for all audiences.

 

[abdo123]

I'm a biologist too and I've been reading JAES papers for a few decades. I don't agree. As with even the most reputable bioscience journals, the actual quality of work varies , and certain 'wrong' results get published, but I don't see the phenomena you cite.

Like i said i cannot comment on the content itself, just the way it was written. And yeah it’s written like shit almost all the time because that journal has no standards.

 

[voodooless]

No, I am just commenting on "difficulty" of embedding data in random bits. I have not looked to see what MQA does.

While you are correct with your RF analogy, I think I have an easier to understand one: right now we're actually all looking at random bit's being transported over the internet from the ASR webserver to our browsers. It's all encrypted, and those data streams are in essence pure noise, still, you can read this, can't you? An MQA stream is equivalent in that sense. It's just that the "encryption" scheme is far less elaborate than what is used to bring you this text. Counter to that, greater care has been taken to shape the random noise distribution into the high frequencies to make it less obvious. In RF they use something similar to get uniform power distribution called data whitening. This essentially converts any bitstream into white noise. But don't confuse this with what Amir explained in his example, it's again another principle I'm not saying that MQA uses this technique specifically. There are other things to achieve similar results, and you'd still need to shape the result. It's just an illustration of how you can make data look seemingly random. @mansr might know exactly what is being used.

You can do lot's of fun things with a pseudo-random generator as well. The kicker is that it is predictable, as long as you know the seed and number of iterations. This means that if you have a clock that poops out a pseudo-random number every second, you can predict the next number until eternity. You can use this for all kinds of clever schemes. So as you can see: there are many methods to generate a seemingly random signal that hides actual data. And most of them are very efficient as well: for every random byte, you get one byte of actual data back.

BTW, could be a fun experiment to take an MQA file, chop off the upper 17 bits, amplify what is left and then listen to it to see how it sounds. Especially for young people.

 

[firedog]

A few years back, I remember reading Warner was planning to convert its entire catalog to MQA. There where similar statements from Universal and Sony but these two seem to have pulled back. Also, it seems Warner has slowed down a bit -- I'm not sure what can be inferred by that. Perhaps non-MQA streaming is safe (as in it will continue to exist) -- but it did not seem like that back then.

I'm not sure how you got to this conclusion. Warners did a mass release of MQA titles over the past year and eliminated non-MQA from their Tidal catalog for those titles. If they have slowed down in the very recent past, I'd guess it's only till the next MQA "dump".

But maybe the lack of MQA at Apple, Amazon, and Spotify lossless/hi-res will be the death knell for MQA. Tidal is an insignificant player in the broader market, and if MQA is only readily available at Tidal, it won't have long term viability.