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

[ebslo]

I don't understand your question. MQA encodes all the music information it can identify above 22.1 kHz. So in 88 kHz sampling, it would find and encode music content to 44.1 kHz bandwidth. In that regard from musical information point of view, it is the same as FLAC at 88 kHz. It is also the same as 176 kHz because there is nothing else left to encode from 44 to 88 kHz.

Thank you, even though I failed to effectively word my question, you have clarified the point.

 

[q3cpma]

OP did do something useful. He did more than that actually by clearly showing that MQA is not mathematically lossless so no one should remotely assume that.

The rest of your post I don't understand at all. The blog post is full of technical information which I have been using to shine more light on the topic. It is not remotely "marketing bullshit." Here is an example:

View attachment 132762

This is clear, technical information that pokes a massive hole in the technical relevance of OP's test. There is noting marketing here. It says the worse case musical information they have found out of encoding millions of tracks shows the spectrum to be well less than OP's test clips. It is authoritative, data based, and shows significant depth of knowledge. This makes your characterization completely wrong and indicative of not reading what you are talking about.

I was talking about https://www.bobtalks.co.uk/a-deeper-look/all-that-glitters-is-not-golden/ if there's a misunderstanding. I won't do each point in detail, but I pointed out that (4) is bullshit that doesn't consider electronic music, a problem solved by actually lossless solutions because they don't have "killer samples". Or (2) that was discussed at length: MQA has never been "clear" about the lossy status of their codec. Or I could talk about their "blur" thing and how they're keen to talk about psychoacoustic without providing any proof that it is an audible problem in non degenerate cases.

Honestly, I don't have an enormous beef against MQA, it's just so obvious that it's a money scheme to me that I simply don't understand why it must get special consideration. My position is just that it doesn't solve anything real over Redbook and good lossy codecs when bandwidth problems actually occur.

 

[DimitryZ]

If it's aiming to be "perceptually lossless", why not compare it with lossy codecs with the same goal, instead? Because it's trying to be inbetween lossless and lossy? Maybe lossywav or wavpack's lossy mode, then. Another problem that is solved by real lossless is that the "made for real music and not test tones" bullshit doesn't include electronic music that could have both an inexistent noise floor (digital silence + possible dither) and some "test tone" like components.


I kinda agree that the OP lacked some knowledge, and he did ran way too far with his suspicions (even if they may be partly shared by others, at least as concern) but he nonetheless did something useful. But I can't fathom how you could call the scandalous marketing bullshit full of rhetoric tricks used in the blog post response as just an "explanation".

It's not perceptually lossless. It's lossless to your equipment's limits to reproduce loss.

I guess you can say it's perceptually lossless to your equipment, which makes it totally lossless to you.

 

[q3cpma]

It's not perceptually lossless. It's lossless to your equipment's limits to reproduce loss.

So, does it preserve digital silence in electronic music even if 3 LSB are converted to noise (in the 16 bit version)?

 

[amirm]

This is of course hypothetical, but let's say the source file is 24-bit/96kHz PCM. Let's also say #2 is competently downsampled from that same source.

OK, I thought you were asking about the source and MQA being the same rates.

I ran an experiment with 44 and 88 kHz sampling since this is what I downloaded from 2L site. The result is close:

The "converted" file is 24 bit Flac at 44.1 kHz at 42 megabytes, derived from 88 kHz original file. MQA version at 24/44.1 kHz is a bit larger at 45 Megabytes (it was derived from higher sample rate file though than 88 kHz). I don't have an MQA encoder to encode the 88 kHz file so can't quite simulate what you are asking.

 

[DimitryZ]

So, does it preserve digital silence in electronic music even if 3 LSB are converted to noise (in the 16 bit version)?

I think so. If there is digital silence for some amount of time there is no MQA content for that duration either.

 

[amirm]

If it's aiming to be "perceptually lossless", why not compare it with lossy codecs with the same goal, instead?

I already explained MQA is NOT perceptually lossless. There is no psychoacoustic model in it to determine what is or is not audible. Instead, it attempts to find useful musical information in the content and delivers those bits and not much more. The market would not accept a lossy encoder that is perceptually based since by definition such a codec would chop everything above 20 kHz, giving you less than CD even! In practice, lossy high res codecs do deliver some ultrasonics but not what was in the original file.

 

[q3cpma]

I think so. If there is digital silence for some amount of time there is no MQA content for that duration either.

OP's test shows some stuff added in the silent segment of the "RMAA test sequence" (one channel only), at least.

 

[q3cpma]

I already explained MQA is NOT perceptually lossless. There is no psychoacoustic model in it to determine what is or is not audible. Instead, it attempts to find useful musical information in the content and delivers those bits and not much more. The market would not accept a lossy encoder that is perceptually based since by definition such a codec would chop everything above 20 kHz, giving you less than CD even! In practice, lossy high res codecs do deliver some ultrasonics but not what was in the original file.

Don't know, that "useful musical information" bit must follow a model that I'd certainly qualify as backed by psychoacoustics. But I literally don't know, as my interest isn't great, so please excuse me if this was already explained at length.

 

[PierreV]

Amir, single handedly defending MQA vs hordes of ignorants. (ca 2021)

 

[Jimbob54]

Does the new middle tier deliver MQA but block decoding? That's evil.

That's the fear. We still need a friendly Aussie to do some work. I don't know what proportion of the Tidal catalog that maxed out at redbook previously has been replaced by mqa (believe some Warners stuff has) and what has a standalone redbook file as well as the mqa higher res one.

 

[amirm]

Don't know, that "useful musical information" bit must follow a model that I'd certainly qualify as backed by psychoacoustics.

No, it just follows signal processing that says what is the quantization noise and what needs to be encoded.

 

[earlevel]

- Apple Lossless first needs to prove it's really lossless and not just "perceptually" lossless, ie. same thing as before but with higher sample rates

Apple's press release:

Apple Music will also make its catalog of more than 75 million songs available in Lossless Audio. Apple uses ALAC (Apple Lossless Audio Codec) to preserve every single bit of the original audio file. This means Apple Music subscribers will be able to hear the exact same thing that the artists created in the studio.

In addition to that, Apple has always been clear what they mean by "lossless", they are computer makers (that write software to sell their computers), they are not music marketers. ALAC has always been lossless in the programmer's—not the listener's—sense. Just saying I'd be happy to take all bets.

 

[amirm]

That's the fear. We still need a friendly Aussie to do some work. I don't know what proportion of the Tidal catalog that maxed out at redbook previously has been replaced by mqa (believe some Warners stuff has) and what has a standalone redbook file as well as the mqa higher res one.

It would be trivial to fix that issue then if it comes to pass. Person gets music from Tidal that is MQA encoded in that tier and won't play. Direct complaint to Tidal will either get that file replaced or make MQA decoding part of that tier.

 

[q3cpma]

No, it just follows signal processing that says what is the quantization noise and what needs to be encoded.

But it doesn't just "truncate" one LSB, which is what regular TPDF dither adds, no?

 

[DimitryZ]

OP's test shows some stuff added in the silent segment of the "RMAA test sequence" (one channel only), at least.

GO's test was fatally flawed, since he never did his homework and learned how the system operates. The only thing he proved is that he is very bad at this stuff. Embarrassing, really.

 

[Jimbob54]

It would be trivial to fix that issue then if it comes to pass. Person gets music from Tidal that is MQA encoded in that tier and won't play. Direct complaint to Tidal will either get that file replaced or make MQA decoding part of that tier.

It will play. That's one of the main selling points isn't it? Plays as (nearly) lossless on anything, or core decode on relevant software/ hardware or full 57 origami folds on fully compatible hardware.

But folks are questioning what an undecoded mqa file is. Apparently less than redbook.

 

[tmtomh]

Thank you!

The answer to your first question is yes, largely due to their "deblurring" and lowering the noise in older recordings. My post only dealt with the accusation of LOSSY, which is meaningless, once the loss is below audibility.

The answer to your second question is yes, but for only one of your reasons - slight sonic degradation for people who don't have a decoder AND can hear above 16KHz. The rest of your reasons only become relevant if MQA takes over the world and that's clearly not happening.

Thanks for your reply.

I'm sorry, but the answer to my first question - does MQA provide anything of value beyond PCM - is most certainly not "yes," since there is no evidence that MQA's deblurring (a) works, (b) is necessary, and (c) is even possible with any recording that contains sources recorded through more than one ADC. As for lowering the noise floor in older recordings, I have no idea what you're referring to. No digital sampling or processing algorithm (except digital noise reduction or treble-lowering EQ) lowers the noise floor of analogue recordings, and plain old 16-bit PCM with no noise-shaping dither already has a noise floor lower than any analogue recording.

Similarly, I would respectfully push back RE the question of whether MQA does anything actively negative.

On the "slight sonic degradation" of undecoded MQA point, I don't think it's about hearing above 16kHz or any particular frequency, because frequency reproduction is about sample rate. It's about bit depth and the noise-floor issues that go with it. The MQA encoding process puts non-random noise/information in the lower bits, which compromises/reduces the bit depth.

Or am I misunderstanding and are you saying that MQA's leaky filters don't really matter unless you can hear above 16kHz? If that's what you are saying, then I understand your logic but I would ask if the aliasing produced by leaky filters (and the potential IMD produced by the presence of those unwanted aliased frequencies) only impacts the very high end of the audible spectrum. I'm not sure about that.

As for my other two points about MQA creating problems by increasing the costs of DACs and in some cases forcing MQA reconstruction filters to be applied to all content played through a DAC, you are certainly correct that consumers can avoid that simply by purchasing non-MQA DACs. However, I would argue that your position in this regard is fairly weak, since the adoption rate of MQA by DAC makers has been far, far higher than the adoption rate of MQA by streaming services (and digital-file vendors). So one's choice in DAC hardware is already notably reduced by MQA, even if one's choice in music streaming is not (given the advent of Apple lossless/HD and Amazon HD, among others).

 

[ebslo]

OK, I thought you were asking about the source and MQA being the same rates.

I ran an experiment with 44 and 88 kHz sampling since this is what I downloaded from 2L site. The result is close:

View attachment 132764

The "converted" file is 24 bit Flac at 44.1 kHz at 42 megabytes, derived from 88 kHz original file. MQA version at 24/44.1 kHz is a bit larger at 45 Megabytes (it was derived from higher sample rate file though than 88 kHz). I don't have an MQA encoder to encode the 88 kHz file so can't quite simulate what you are asking.

Thank you very much! That gives me effectively the answer I was looking for, as you already explained that the MQA encoder output will be essentially the same size regardless of whether the input was 88kHz or something higher (unless I misinterpreted something).

Reposting my original question to @abdo123 below with full context, I was addressing his assertion that MQA files were smaller for equivalent perceptually relevant information, and you have answered the size part of that in the negative. The perceptual equivalence part is of course much more difficult to prove definitively as evidenced by the page count of this thread, but over the course of it I think there is enough to have reached an at least somewhat informed opinion.

and when MQA says they're better than lossless they are indeed better (not necessarily to the consumers) because they're smaller and easier to maintain on servers and give perceptually the same thing anyway.

I'm still unclear on this point. Could you please clarify the following?
1. Is MQA encoded to 24-bit/48kHz smaller than FLAC-encoded 24-bit/48kHz PCM?
2. Does MQA encoded to 24-bit/48kHz deliver perceptually the same, more, or less than FLAC-encoded 24-bit/48kHz PCM?

 

[levimax]

Let's see a few ABX test outputs on who can pass it and what they are hearing different. Don't wait on me on that. You don't need me for any of this type of testing.

This to me is THE PROBLEM. If Tidal had left up the original versions of the Warner Catalog that they replaced with MQA versions then I could ABX the two versions and decide for myself if I could hear any difference and if I could if I had any preference. The fact that the original versions were removed after the MQA versions were put up combined with MQA's refusal to publish "before and after MQA demonstration tracks", while "proving" nothing certainly does not pass an ignorant layman's "smell test". If I had a super codec that "sounds better" I would want to prove it to the world and shout it from the rooftops. With no directly comparable versions on Tidal, no directly comparable versions from MQA, and no access to an MQA encoder that leaves only one alternative to find out what MQA really is about which is what GoldenOne did... it was not perfect but it is a start.