If I keep taking her out so I can hide, she'll never get fully frozen.
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
MQA Deep Dive - I published music on tidal to test MQA
- Thread starter GoldenOne
- Start date
- Status
If I keep taking her out so I can hide, she'll never get fully frozen.
- Joined
- Oct 25, 2019
- Messages
- 11,115
- Likes
- 14,779
If you keep putting her back in she'll never be ripe like a Georgia peach!
Such a delicate balance, but worth it.
Maybe try pickling her ? Thank me later ..
Oh bath tub and 6 gallons of vinegar, peppercorns and onion optional..
Outstanding.
I'll save you some. Maybe try it with some Fava beans and a nice Chianti.
I think you completely missed the point here. He’s alluding to the usage of the term “lossless” in the MQA FAQ, which I think most people here find rather disingenuous at best, but Amir just said: well they just have a different definition, so what’s the problem (heavily paraphrasing here
)…You can still disagree with the choice of words, tone and message, but if you do at least do it the the right context.
Who said it was?
- Joined
- Oct 25, 2019
- Messages
- 11,115
- Likes
- 14,779
Anyway. Pineapple
Yep. In my discipline, conference presentations on research are more often preliminary sharing of what you're working on. In order to create a conversation.
But if you share the full research, your analysis, and conclusions, then no good journal is going to take it. So you simply wouldn't do what some of these AES papers are doing unless you didn't feel like the work was significant enough to make it into a good journal.
Last edited:
- Joined
- May 28, 2021
- Messages
- 32
- Likes
- 53
I don't think I'm confused. Let's retrace this strand of the thread:
comment I was responding to:
Here I understand fully that they are referring to mathematically lossless compression all of us love so much, e.g. FLAC.
my response:
to which, the commenter of interest responds:
I don't think I'm confused here. I'm clearly addressing the fact that most audiophiles mean mathematically lossless compression when using the term "lossless". I'm trying to start a discussion on this matter and of what value we, as lay people, should put in these concepts and understanding them as they pertain to our hobby.
The only (incredibly unfounded and rude) allusion here is :
This last part I do apologize for. Poorly placed on my part.
I meant that in address to everyone not just the commenter. I was trying to reiterate the fact that I (and I'm assuming many other audiophiles) have a pretty skewed experience with the term "lossless". 99.999999% of my usage of that word has been with regards to audio.
So ultimately, I am seeing that there is much more understanding needed of the concept of "lossless" within the community, and that is where I'd like some serious discussion to head... I want to learn.
What I don't want is a bunch of anonymous snarks insinuating Amir's lack of expertise.
comment I was responding to:
Here I understand fully that they are referring to mathematically lossless compression all of us love so much, e.g. FLAC.
my response:
to which, the commenter of interest responds:
I don't think I'm confused here. I'm clearly addressing the fact that most audiophiles mean mathematically lossless compression when using the term "lossless". I'm trying to start a discussion on this matter and of what value we, as lay people, should put in these concepts and understanding them as they pertain to our hobby.
The only (incredibly unfounded and rude) allusion here is :
What value is this adding to the conversation aside from more unfounded personal attacks?
This last part I do apologize for. Poorly placed on my part.
I meant that in address to everyone not just the commenter. I was trying to reiterate the fact that I (and I'm assuming many other audiophiles) have a pretty skewed experience with the term "lossless". 99.999999% of my usage of that word has been with regards to audio.
So ultimately, I am seeing that there is much more understanding needed of the concept of "lossless" within the community, and that is where I'd like some serious discussion to head... I want to learn.
What I don't want is a bunch of anonymous snarks insinuating Amir's lack of expertise.
Last edited:
You are not confused regarding the whole picture I think, so don’t worry
. However regarding the term “lossless” in audio: I think it has been quite clear for years what is meant by that. Yes, you can assign different meanings, but that just confuses things. A tactic that i think is used by MQA on purpose. I’m all for exploring the various definitions of “lossless” though. It’s an interesting topic for sure.
Zensō
Major Contributor
We already have a term for “perceptually lossless”, which I believe is “transparent”, no?
Raindog123
Major Contributor
I may have contributed to a misunderstanding about different types of lossless encoding so let me say it succinctly outside of a conversation:
1. Mathematically lossless. This type of compression produces identical bits that was fed to the encoder, at the output of the decoder. It doesn't matter what those bits are with respect to their information content. Lossless encoders are trained on the likely content type (e.g. music versus computer programs), but otherwise are dumb in that they don't apply any psychoacoustics to what they are encoding. If music fades to noise, they will happily encode that noise.
I wrote a brief tutorial on mathematically lossless compression a few years ago. Read it for a bit more information on the topic: https://www.audiosciencereview.com/forum/index.php?threads/lossless-audio-compression.17/
When people hear the term "lossless," this is the type they assume.
Typical compression ratios are 3:1 or so.
2. Perceptually lossless. This is a lossy encoder that is told to keep quality at 100%. It uses psychoacoustic model of hearing and lossless encoding technique to reduce the file size. You can get about 50% more compression efficiency than mathematically lossless encoding (nearly half the size). Data rate would be in 400 kbps range and variable (just like mathematically lossless).
Interestingly most lossy codecs also support high sample rates. AAC, WMA Pro, Dolby AC-3 are examples. I say interestingly because any psychoacoustic model would say chop off everything above 20 kHz so encoding wider bandwidth violates this. The standard technique is to extend the model to have an exponential drop off that correlates with what comes before 20 kHz. Note that you can code high-res data in a lossy manner in perceptually lossless mode or not.
With respect to bit depth, that aspect of high-resolution audio is better represented so there is some value in extending a lossy codec to beyond CD spec of 16/44.1.
3. MQA (or really MQL as they are calling it now). This type of compression is different in that it only codes music above the baseband 44.1/48 kHz based on information/noise level it has. If its max range is 6 dB above noise floor, then all it takes is 1 bit to represent it, not 24. To do this, an analysis of the content must be made in the encoder and decisions made about what is noise and what is content above or below noise.
There is no psychoacoustic model here as again, such a model would say to not encode above 20 kHz. Instead, there is implicit assumption that music's spectrum/information content drops exponentially as frequencies increase and hence becomes easier and easier to encode it (use fewer bits), to describe it. MQA says let's allocate the fewest bits we need to describe the spectrum above the baseband. If the music needs more bits, it is given. If not, it is not.
There is a tricky complication here. Instead of creating a new format, MQA decided to embed the extra information it needs for the spectrum above baseband (44.1/48 kHz sampling), in the data there. The data is randomized so it raises the noise level there in high frequencies. So if you don't decode MQA, you are degrading the baseband some. This backward compatibility is an added feature but unrelated to how the encoding itself works.
This type of encoding is NOT perceptually lossless. It is a completely different animal as it attempts to fully preserve the musical information with no regards to whether some of it could have been thrown out due to inaudibility. To wit, its file size can never be smaller than the baseband (44.1/48 kHz) sampling whereas a perceptually lossless encoding can produce smaller file sizes. So it is a less efficient method of coding than perceptually lossless but somewhat better than mathematically lossless.
NOTE: we focus a lot on spectrum here but MQA also attempts to deliver more than 16 bits. There, we can't just say "it is above audible band so who cares."
1. Mathematically lossless. This type of compression produces identical bits that was fed to the encoder, at the output of the decoder. It doesn't matter what those bits are with respect to their information content. Lossless encoders are trained on the likely content type (e.g. music versus computer programs), but otherwise are dumb in that they don't apply any psychoacoustics to what they are encoding. If music fades to noise, they will happily encode that noise.
I wrote a brief tutorial on mathematically lossless compression a few years ago. Read it for a bit more information on the topic: https://www.audiosciencereview.com/forum/index.php?threads/lossless-audio-compression.17/
When people hear the term "lossless," this is the type they assume.
Typical compression ratios are 3:1 or so.
2. Perceptually lossless. This is a lossy encoder that is told to keep quality at 100%. It uses psychoacoustic model of hearing and lossless encoding technique to reduce the file size. You can get about 50% more compression efficiency than mathematically lossless encoding (nearly half the size). Data rate would be in 400 kbps range and variable (just like mathematically lossless).
Interestingly most lossy codecs also support high sample rates. AAC, WMA Pro, Dolby AC-3 are examples. I say interestingly because any psychoacoustic model would say chop off everything above 20 kHz so encoding wider bandwidth violates this. The standard technique is to extend the model to have an exponential drop off that correlates with what comes before 20 kHz. Note that you can code high-res data in a lossy manner in perceptually lossless mode or not.
With respect to bit depth, that aspect of high-resolution audio is better represented so there is some value in extending a lossy codec to beyond CD spec of 16/44.1.
3. MQA (or really MQL as they are calling it now). This type of compression is different in that it only codes music above the baseband 44.1/48 kHz based on information/noise level it has. If its max range is 6 dB above noise floor, then all it takes is 1 bit to represent it, not 24. To do this, an analysis of the content must be made in the encoder and decisions made about what is noise and what is content above or below noise.
There is no psychoacoustic model here as again, such a model would say to not encode above 20 kHz. Instead, there is implicit assumption that music's spectrum/information content drops exponentially as frequencies increase and hence becomes easier and easier to encode it (use fewer bits), to describe it. MQA says let's allocate the fewest bits we need to describe the spectrum above the baseband. If the music needs more bits, it is given. If not, it is not.
There is a tricky complication here. Instead of creating a new format, MQA decided to embed the extra information it needs for the spectrum above baseband (44.1/48 kHz sampling), in the data there. The data is randomized so it raises the noise level there in high frequencies. So if you don't decode MQA, you are degrading the baseband some. This backward compatibility is an added feature but unrelated to how the encoding itself works.
This type of encoding is NOT perceptually lossless. It is a completely different animal as it attempts to fully preserve the musical information with no regards to whether some of it could have been thrown out due to inaudibility. To wit, its file size can never be smaller than the baseband (44.1/48 kHz) sampling whereas a perceptually lossless encoding can produce smaller file sizes. So it is a less efficient method of coding than perceptually lossless but somewhat better than mathematically lossless.
NOTE: we focus a lot on spectrum here but MQA also attempts to deliver more than 16 bits. There, we can't just say "it is above audible band so who cares."
- Joined
- May 28, 2021
- Messages
- 32
- Likes
- 53
You are not confused regarding the whole picture I think, so don’t worry
I agree 100%. And as I've stated before, I don't think MQA provides us with any practical value over something like FLAC in the era of 4k streaming and cheap cloud storage. A whole lot of my local storage of music is in FLAC, and the entire library takes up maybe 5% of my available storage.
Even if unnecessary for everyday consumer, better and newer compression methods are for sure interesting and their claims should be methodically scrutinized for merits. Couple percents off of my music library means jack for me personally but may mean a lot for gigantic databases of audio. Of course, the complaint (that I share) here is music consumers are paying for the cost of that through licensing, marketing etc. but receive very limited end user value.
(this is general statement for everone) I think it's important for us to understand that judging MQA as a technology is completely independent of judging MQA as a profit making venture. Both can exist in great harmony, and in reality so many startups/spin-offs from academia are indeed based on this concept. As a business venture, I'm leaning heavily on the side of "this is going up in smoke in a few years". As for the compression method having merits, I'll patiently wait for more of Professor M's analyses.
In the end, I think if a perceptually lossless compression method is able to save us more space and remain indistinguishable from something like FLAC, then we must arrive at the consensus that it is, as far as performance is concerned, objectively better.
The thing is... how many audiophiles are worried about saving disc space? I'm a data hoarder when it comes to music. I want to possess as much as possible the entirety of the moments in time when the music was experienced. I don't care that it is physically imperceptible to me. As long as I can verify it is there, and that I possess it, my monkey brain will be at ease. I recognize this as a purely materialistic and self-centered endeavor, not too unlike having a super car as garage ornamentation. As far as materialistic ego stroking goes, though, I think ours is a relatively harmless exercise.
Raindog123
Major Contributor
@amirm I do not think MQA is claiming to be “perceptually lossless”. Their position is “MQA coding is lossless [in a traditional, mathematical sense] as far as the source material stays within the [spectral] boundaries of the MQA design (‘the triangle’)”.
(Personally, I would be ok with that, would it be substantiated/proven. And explained what benefit it offers compared to “traditionally lossless (within ‘the rectangle’)“ PCM, but it’s a different story.)
(Personally, I would be ok with that, would it be substantiated/proven. And explained what benefit it offers compared to “traditionally lossless (within ‘the rectangle’)“ PCM, but it’s a different story.)
Last edited:
I think there still is more confusion here
. You talk about the codecs and what they do. But you can also look at the result, independently from the codec. That changes the meaning of the second one especially, and would let MQA be perceptually lossless as well (as in: you perceive the file just as you would the original.. which is what MQA is advocating anyway).. the ultrasonics you get as a bonus .. as I said.. more confusion 
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).If my understanding is correct, while interesting, MQA doesn't seem to solve any problem in our 4k streaming/cloud storage age. (Yes, I'm slow.. thank for being patient with us smooth brained folks
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.
Back to compression, we need to keep in mind that music distribution is horrible business. Steve Jobs destroyed the business model by taking out most of the profit out of the music distribution, and putting all the profit in the hardware to sell (iPod). A track that you sell for 99 cents, may incur royalties of $90 cents leaving you almost no money to make a living from. With CDs, they cost 50 cents to make and they sell them for $13, leaving plenty of money for everyone to make a living from the content owner to record shop that used to sell them.
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.
The above is why Spotify bleeds so much cash (over $2M/day!). This is why Tidal is losing money left and right. This is why hardly any other distribution company has lasted. In their stupidity, record labels have created an environment where it is not profitable to distribute their content so they are stuck with immense power in the hands of technology companies.
Long way of saying
, saving some cost in bandwidth was a useful selling tool for MQA with respect to Tidal. They got to have "high-res" offering at lower cost than if they had shipped Flac files. Keep in mind that Apple and Amazon have likely much lower bandwidth costs than Tidal since they use so much of it. As such, the savings from MQA is not as material to them as it is to Tidal.- Joined
- May 28, 2021
- Messages
- 32
- Likes
- 53
@amirm
Thank you so much, professor. I'm sure you've discussed all these points before in videos and other threads so I really appreciate you reiterating everything for those of us who find it hard to sift through all the relevant information in the forum as well as on your videos
(that last bit is just me, but I'm using "some of us" to shield me from the shame)
So: MQL's claims are that it achieves the results of mathematical lossless through a novel process which produces slightly more efficient compression.
Thank you so much, professor. I'm sure you've discussed all these points before in videos and other threads so I really appreciate you reiterating everything for those of us who find it hard to sift through all the relevant information in the forum as well as on your videos
(that last bit is just me, but I'm using "some of us" to shield me from the shame
)So: MQL's claims are that it achieves the results of mathematical lossless through a novel process which produces slightly more efficient compression.
I’m not 100% sure this is correct, at least not for all tracks, but it might be a minority or an artifact: We’ve seen some spectrum early on in the thread of a core decode that showed a whole lot of aliasing, exactly the way one would expect from the type of interpolation/reconstruction filter. The non-MQA version did not have this.
That is why I’m very interested in your test track with the brickwall filter blinky thing
- Joined
- May 28, 2021
- Messages
- 32
- Likes
- 53
- Status
Similar threads
