Friday 14 October 2016

MUSINGS: Keeping it simple... MQA is a partially lossy CODEC.

Connoisseurs of French cuisine will know this guy...
In the last few days, I noticed my inbox containing a number of questions again about MQA... Presumably due to a combination of the last blog post with Agitater's comments and I guess some news on MQA after the recent Rocky Mountain Audio Fest 2016 where a number of companies were playing MQA and apparently allowing A/B comparisons of sorts.

I say "of sorts" above because it seems from descriptions the best comparisons were more along the lines of using the same file but with MQA decoding turned on/off "on the fly". Not an actual native 16/44 CD resolution file vs. MQA decoding in a volume controlled state using the same source mastering of course. Remember, the processing itself likely affects the sound quality (indeed, Meridian/MQA wants us to believe MQA processing will still benefit playback even without a decoder and folks testify about this as well). My little ABX test suggests small differences as well.

So, let's just keep it "simple" ("but not simpler" as per Einstein). Here's the bottom line for us consumers IMO...

Bottom line: In the eyes of a consumer, MQA is a file type with partial lossy compression that you have to buy a new DAC for to gain full benefits. [Update: With the introduction of software decoding like in TIDAL which decodes up to 24/96, buying a DAC is not absolutely necessary any more since January 2017 (see here).]

To me, that's the heart of it. Everything else revolves around that statement as we try to justify or criticize this central fact.

The reason why this fact can be difficult to reckon with is because it's a proprietary partially lossy "hi-res" CODEC (COder/DECoder)**. Imagine a 24-bit WAV/AIFF file with a portion containing lossy data (think MP3 or AAC or Opus) embedded in the 24-bit noise floor. The "baseband" sound material up to 44.1/48kHz sampling rate (that is, up to 22.05/24kHz audio frequency) is basically a WAV/AIFF PCM stream down to around the 16th bit. And there's stuff below that 16th bit or so that's encoded in a lossy fashion used to reconstruct frequencies above 22.05/24kHz. With lossless compression used on that 24-bit WAV/AIFF data, it looks like any other FLAC/APE/WV/ALAC file.

Implicit in the "lossyness" is the fact that if we feed a very high resolution file into the encoder - say a DXD 24/352.8 file, and it needs to be squished into a 24/44 container, it just means more data is thrown away; at least the portion in the ultrasonic spectrum the algorithm deems as not worth keeping (presumably based on their concept of the spectral content of music). Sure, when you play back the MQA file, it might be reconstituted into a 24/352.8 stream and this looks reassuring with your DAC displaying "352.8kHz" if supported since that was the "original resolution", but in reality, we have to accept that relatively more data was lost compared to a 24/88 track compressed into that same 24/44 data stream! If you look at the 2L music store, you'll see the MQA files listed as "Original Resolution" even though of course the file size would be equivalent to 24/44 or 24/48.

Like all CODECs, all we really need is a decoder to listen with full quality. In foobar2000 for example, we already have decoder plugins for DSD/SACD, DTS, MLP, etc... Not to forget the usual audio formats like FLAC, WV, ALAC, APE, etc... This list includes HDCD which is of course a similar idea; that embedded in an otherwise "lossless" stream within the least significant, likely inaudible bit, data can be recovered to enhance the overall potential quality.

If Meridian/MQA wanted to, there's really no reason why they can't release a plugin for foobar that will decode those files (like these 2L demos for example) and allow reconstruction and playback at 24/192 or whatever samplerate we might choose. This is basically what the firmware upgrade in a DAC does; they added a software component that internally recognizes the MQA stream being fed and reconstructs the lossless + lossy parts as an upsampled PCM stream that feeds the DAC chip(s) up to the maximum resolution the DAC accepts. Of course the engineers spent a good amount of time and effort doing this. I'm sure it's not easy, good for them! But conceptually it's no different than something like a DTS/Dolby Digital decoder accepting a bitstream and creating a final PCM output from this. Sure, for streaming applications, a "high-rezzy" stream at 24/44 bitrate is reasonable when bandwidth is limited but still about 50% greater than standard CD resolution. But if you're not into streaming, and can download a full, unadulterated, 24/192 "studio master", would you actually prefer the 24/48 MQA compressed version? I can't imagine many audiophiles would if they took some time to actually think about this!

The frustration and disbelief for many are claims that it's "more" than just a CODEC. It's the fancy talk about how this is all done (the "origami" process) and why it's supposedly good (improved time domain accuracy thru the "de-blur" DSP algorithm). Over the last couple of years since the fancy release party, it's this impressionistic marketing chatter along with hyperbolic ("revolutionary" anyone?) claims and testimony that IMO fuels criticism. The claims are hard to accept, and the testimonials from the usual audiophile press suspects are even harder to swallow.

For me, there are "simply" two major issues I find wrapped up in what otherwise is a proprietary CODEC that makes this whole scheme debatable.

1. The "de-blur" algorithm supposedly makes it sound better:
Supposedly data about the impulse responses of the studio equipment fed into the encoder can lead to a more accurately time-aligned version of the music. Supposedly, based on magazine testimony, this audible difference is remarkable. Now suppose I take this at face value and accept that indeed it sounds better, as an audiophile who cares about quality and wants the best sounding version I can get, I'd be wondering then why this de-blur algorithm cannot be applied to a fully lossless 24/96 or 24/192 download? Wouldn't that potentially be even better!? Wouldn't I ultimately want that instead of yet another intermediate "format" with lossy compression? There's no reason to marry this de-blur processing with the partially lossy CODEC which my DACs can't decode. IMO this is no different than any of the myriad studio DSP plugins used in the production chain that can potentially make the music sound "better". This is why I've always felt that there needs to be a dissociation between the CODEC piece and the "de-blur" DSP piece as I suggested months ago. By all means, provide the de-blur processing as a tool on the production side for the artists and sound engineers so they can apply it to anything from CD-quality to DXD.
[Let's not forget that the significant time-domain errors inherent in speaker systems during playback cannot be cured by a universal de-blur algorithm in any event as discuss here.]

By the way, I see claims that a good portion of the Warner library has been converted to MQA. (According to Dr. AIX, it's only the already-digitized "hi-res" albums - presumably the same ones we can already buy as downloads these days.) Hmmm, if indeed 3500 albums have already been done as reported, who honestly believes they did this with all the care needed to input the time domain data to produce the best possible de-blurred output? Are all these albums "authenticated" - whatever that really means? This is all also assuming genuine "de-blurring" is even possible given the complexity of modern studio production!

2. It's a proprietary "format":
We've been very fortunate as audiophiles. All these years we've enjoyed openly compatible PCM and DSD files. In computer audio, all sorts of great tools have been made available for encoding, decoding, meta-data tagging, and format conversions. Who wants certain music files now wrapped up in a proprietary mechanism such that you would not be able to experience the full quality unless you buy specific hardware incorporating a license through a single third party? I trust the likely answer is nobody except those who can gain something - likely financially. If we are to voluntarily submit to a loss of freedom (ie. the freedom to pick whatever DAC we want to take full advantage of the files we purchase or data we stream), there better be a great reason!

The point is not that I refuse to buy new hardware. DACs break down and need to be replaced, we might want to treat ourselves with new gear. No problem. My opinion is that if MQA has so much faith in their overall process, they should release a computer software decoder so people will start purchasing the MQA files such that in time, those people can be enticed to buy MQA-enabled hardware because they've built up a library of convenient (smaller), good sounding files. The market will decide based on demand for playback. Things appear to be totally backwards at this point! Such and such hardware manufacturer announces MQA compatibility, but seriously, where's the software? As audiophiles, we are often reminded to put the music first because that's the goal of this hobby after all... Perhaps the industry should lead by example?

Since I last wrote about MQA in July, I'm glad there is more A/B-type listening being done at the audio shows even if rather limited. That's nice. But because this is software, they really should either release a software decoder or consider at least releasing an original 24/192 source file and the decoded 24/192 version having gone through the MQA process for audiophiles to audition the difference in quality for themselves. This is obviously the best way to allow others to experience the difference MQA could make (again, if they truly believe the difference is that remarkable).

Note that I don't expect they would do such a thing. They don't have to, and there could be business reasons not to. I can also imagine why they won't do this based on other peripheral rationale. MQA/Meridian could say things like: "Unless the DAC is authenticated it won't be an accurate evaluation..." "Your DAC will mess up the time domain accuracy..." I will leave you, the readers to mull over the validity of such statements.

Just in case I come across as too critical, too quickly against MQA, realize that I have nothing against MQA or Meridian. I've always enjoyed Meridian's hardware when demoed at the local dealer, and I thought the MLP (Meridian Lossless Packing) CODEC used back in the days of DVD-A and still used as Dolby TrueHD audio on Blu-rays is a great lossless compression technique. I have no connection to them whatsoever which also allows me to be free to speak my mind without jeopardizing any relationships. In many ways, this saga of MQA reminds me of the whole Pono embarrassment and all those claims Neil Young made if we roll back the clock a couple of years. The likelihood of longterm commercial success of high-res in a Toblerone-shaped portable digital audio player with a "me too" online music store was tenuous then, and the likelihood of acceptance for "high-rezzy" streaming through MQA is highly questionable now. I frankly fail to see the consumer at large interested in a bump in streaming bandwidth (~50% beyond Tidal's standard lossless CD-quality FLAC) along with a not-quite-lossless "high-res" audio CODEC. As others have noted, this is a solution for a problem nobody really knew we had.

Looking at the writings of the audiophile press at least here in North America, has any writer demonstrated any criticism of MQA or penned a thoughtful opinion piece questioning the wisdom or claims? I sometimes wonder... If in this hobby there is no such thing as "freedom of the press" to investigate, question, even criticize, what becomes of the hobby? What happens to real innovation propelled by feedback and frank discussion? Where is leadership to advocate for things that matter - like maybe a relentless push to end the Loudness Wars or at least lobby for release of "uncompressed" masterings by the major studios?

These days, at least with the articles I see about MQA, there seems to be little beyond one big advertising spin by the "pros".

Bottom line: In the eyes of a consumer, MQA is a file type with partial lossy compression that you have to buy a new DAC for to gain full benefits. And claims of qualitative improvement based on various purported benefits still need to be substantiated.

** Regarding the lossy nature, in this series of articles in Stereophile, there was a quote from Bob Stuart:
"• b) There is no foolery here: MQA does indeed reconstruct a remarkably close approximation to the original ultrasonic information from the lower bits of a 24-bit signal."
(emphasis mine)

I made a comment on this on August 13th after that series was first put up. Interestingly, I can't find that quote again in the current version of the articles on-line. I presume they updated the articles?

--------------- 

Postscript:
As I was finishing the proof-reading, I realized that I ended this post on a rather negative tone. No, it can't end like this :-). Realize that despite the concerns I express above regarding MQA, and the direction of the press / audiophile reporting, I do believe we are living in a Golden Age of sound quality! The computer audio "revolution" piggybacked on the growth of Information Technology has brought us unprecedented access and quality of playback at remarkably low cost (just look at the Raspberry Pi 3/HiFiBerry DAC+ Pro article two weeks ago) whether the folks in the "high end" can admit it or bother to remove personal prejudices when evaluating. Not hard to imagine how this threatens some people in certain segments of the audio world. But for those of us partaking in the "music lover" and "hardware audiophile" hobbies, it's really great times!

Oh, one last thing before I go... Consider giving this album a listen:
BTW... Back to mainly CD purchases for me. Burned by too many "hi-res" downloads with no evidence of better mastering for new music.
For those who enjoy some R&B/jazz/soul and looking for new music (but retro style), I can recommend Nick Waterhouse's new album Never Twice (2016, DR10). Not the best recording / production quality (old Ampex analogue recording gear BTW), much of it I'm sure intentional, but I'm very much enjoying it over the last few nights. He has a couple other excellent albums out also: Holly (2014) and Time's All Gone (2012).

Hope you're all enjoying the music!

PPS (Oct. 15, 2016):
I was reflecting on Jeffrey Barish's comment he left on the post last week. In the world of academics and research that I'm familiar with, it's important to remember that conflict of interest is not defined necessarily by actual conflict. Rather, recognition of perceived conflict of interest is enough cause for concern when it comes to credibility of one's findings and opinions. I hope that the audiophile press recognizes that.

I have heard press apologists state that in a good publication, there is a "wall" between the editorial and advertising sides of the enterprise. Seriously? When the advertisers are essentially the same month to month? The same people end up being friends in trade shows (fly fishing trip!?)? Benefits can be obtained through these relationships whether it's a round of drinks, a nice dinner, or a longterm "loan" of equipment "worth" thousands? Maybe a nice trip to the next product announcement?

I honestly do not see how this "wall" cannot be anything but rather permeable! Look, I'm not pointing fingers at anyone or any publication specifically... Rather as I said above, it's about perception and how that is vigilantly managed. The ability to manage this is the core of being a true professional with ethical obligations not only to the target audience as a journalist, but also to other members of the Industry.

13 comments:

  1. Hey Archimago,
    Excellent article as always. Extra thanks for the mention of Nick Waterhouse. I just sampled some of his tracks over at Amazon and have placed an order for some CDs. Love it when I get turned on to artists I've never heard before.
    All the best.... Carlo.

    ReplyDelete
  2. Hi Archimago:
    You say: "But if you're not into streaming, and can download a full, unadulterated, 24/192 "studio master", would you actually prefer the 24/48 MQA compressed version?"
    I would, if it sounded better than anything else - as MQA and some reviewers claim it does. That claim is the ONLY reason I was interested in MQA. If it was indeed better, the smaller file size and streaming potential would be a slight bonus, but that's all.
    After listening to MQA via an Explorer 2 DAC, I concluded (as I said in another comment on one of your posts) that it was indeed very good. But further listening has convinced me that the same file sounded just as wonderful, when played without MQA decoding through my TEAC UD-501. In fact, everything 2L puts out seems to sound very good, no matter what the format. Specifically, MQA was dazzling with Martin Tingvall's solo piano album, Distance. But the file also sounded spectacular, if slightly different tonally, played without MQA via the TEAC. The fact it that's just a beautifully recorded album.
    My Explorer 2 now sits in a drawer, where it will probably stay. I wouldn't buy further hardware just to play MQA, but wouldn't mind at all if it became a standard software feature of high-quality DACs like the TEAC.
    Cheers, Don

    ReplyDelete
    Replies
    1. Hi Don.

      The *idea* that MQA makes it sound better as per the reviewers' testimonies is important... Perhaps this is why at least for me, from the start, I've been perplexed by the claims in the context of a compression technique which in itself doesn't seem to provide enhanced fidelity.

      How can a partially lossless CODEC make things "sound better"? How can they claim it sounds better than the original 16/44 recording just by running it through the processing?

      It all comes down to that supposed DSP "de-blur" processing. Not the compression format utilizing the "origami", nor IMO the upsampling algorithm and claims of improved impulse response for the digital filtering.

      If the de-blur is truly that worthwhile, then I hope they sell it for studios to use with full hi-res projects and not "force" it as part of this CODEC just so we can access a lossy software/firmware decoder.

      Delete
    2. Isn't improved impulse response for the digital filtering part of the de-blur process?

      Delete
    3. Hi Unknown.

      That's part of the vagueness with the concept of "de-blur". MQA over the years has indeed made that kind of connection but at other times and in other articles, it's denied to be the case.

      I'll write more about the "de-blur" component in an upcoming article.

      Delete
  3. The great thing about MQA is the cool little logos hardware companies can put on their promotional material. It adds value to new products and increases the WOW factor.

    ReplyDelete
  4. This weekend (Dec 10), purely by chance, I walked into a demo of MQA with Meridian powered speakers, Meridian pre, and music from a hard drive supplied by Meridian, with MQA files. This was at a Calgary dealer who had them on loan for a customer to hear. I was not that customer since this stuff is WAY out of my price league. But I got to listen with just one staff member, for about 15 minutes.
    Sampling crow for what I wrote above, I have to say I've never heard anything like it. In fact, after listening to many demos of very expensive stuff over the years, I've never heard a system that approaches the sheer scale, soundstage, reality and musicality of that system. It was emotionally overwhelming, totally engrossing - simply, hauntingly, beautiful.
    I'm not drawing any conclusions about MQA for anyone else here, or even for myself. I'm not even sure of the model number of the speakers, and I wasn't there long. All I can say is that all my tentative earlier conclusions about MQA, and Meridian, are turned on their heard - again.

    ReplyDelete
  5. Are you saying it is possible that the original Tidal HIFI option could sound *better* than the new 'Masters' setting? Considering we can no longer select HIFI setting for albums that have Master option this is... interesting

    ReplyDelete
  6. With TIDAL's announcement, I've started going back and reading much of the discussion I've missed. Not surprisingly, this article asks the same questions I've been pondering as well.

    It wont surprise me if on MQA DACs, something like 352 or 384/24 sounds better, as a lot of DACs use the generic and quite awful filters included. With a good up-sampling algorithm, such as iZotope's, the sound through this DACs can, subjectively, be improved.

    ReplyDelete
  7. "Bottom line: In the eyes of a consumer, MQA is a file type with partial lossy compression that you have to buy a new DAC for to gain full benefits."

    What's partially lossy? :-) It's either lossy or lossless. In this case, the hype cannot restore the audio information lost!!

    ReplyDelete
  8. Experts and marketeers play with and stretch the term "lossy" too much. It is a binary property: either a process is lossless, reversible or it isn't. There are no degrees of lossiness. Even Dr. AIX is wrong as he tries to present resampling and word-length reduction as lossless in his article "Taming the Terminology".

    As Mr. Frank Klemm, the designer of the MPC format, one of the earliest high quality encoders, and proponents of loudness normalization, once stated, "Lossless is not an attribute of a file format, but of a process."

    Examples for "Lossy":
    * Rounding a 24 bit L-PCM file to 16 bit L-PCM. [..]

    If encoding nulls against the original with the difference in the -80 db range, as is typical with the WavPack format, it is of high quality, but still not lossless. Even decoding of an MP3 file to LPCM isn't technically lossless, because the source can technically contain extreme dynamic range, and the frequency components do not sum to round values.

    A format might only be called "lossless" as a convenient short-hand label, because it facilitates lossless editing.

    If the MQA implementation stays compatible with undecoded playback, it sacrifices efficiency, just as other dual-stream formats (DTS MA, WavPack hybrid), and isn't better because of it.

    Call shovel a spade. Compression and encoding, instead of folding and origami.

    ReplyDelete
  9. Archimago a very interesting and informative article, thank you. There currently is a thread on Art Of Sound https://theartofsound.net/forum/showthread.php?61026-so-what-is-your-streaming-preference-and-why initially discussing what people use to stream. However the discussion has turned to the merits of MQA. I quote below my simple thoughts as a music/hifi punter. Also I would add as you rightly point out the proprietery nature of MQA does leave one feeling rather concerned for the future if it establishes itself and is not released into the general domain freeing it completely from DAC's or selected software/streaming suppliers.

    "Ok let’s try and put this into some context. I work with a lot of younger people who think and actually believe that a modern systems such as Google Home, Amazon Echo, Apple HomePod etc actually produce wonderful sound quality, and all they want is 2 so it’s stereo. The answer IMO is for what they do they produce a reasonably good sound but not a true or necessarily an accurate rendition of the original, they do some sudo digital processing to supplement and do what’s been programmed to give the sonic impression of something good. I doubt most of us would consider them an acceptable system to listen to music properly.

    So what’s my point you ask? Well firstly I suspect most of us on here grew up striving for a true and accurate musical rendition from our systems, and actually believe this what most people would want. Secondly I think we probably believe(d) that the music industry (I include hifi manufactures, music producers, and music companies) want the same. Well I suspect that in both cases the answer is no, the reason I say this is that probably for the majority of listeners it is just casual listening and how accurate does not really matter, and to answer the second most manufacturers are about volume sales and quantity, not necessarily quality, whether perceived or real, and what gives most profit.

    So what do we have here in MQA, something that is actually reproducing what was originally recorded, or a method/system that reproduces and enhances what the original was to give a perceived better experience? It’s a very tricky question to answer I think.

    Obviously the same arguments can be applied to other recording methods and reproduction."

    ReplyDelete