MQA vs HiRez: an apples-to-apples comparison - FINAL

[manisandher]

On 10/29/2018 at 6:42 AM, church_mouse said:

... my conclusion was A and C similar, but I preferred A, with B definitely different but not in good way.  The lower end on B was definitely fuller, but the start and stop off the notes seemed to have gone, the thwack on the string missing, and the top end seemed slightly off.

 

Sorry c_m, I forgot to comment on your post. It seems you're hearing things in pretty much the same way I am.

 

On 10/29/2018 at 6:42 AM, church_mouse said:

I now discover that, as with your previous challenge, I liked the Hi-Res least!

 

That makes 3 of us.

 

On 10/29/2018 at 6:42 AM, church_mouse said:

Grasping for a silver-lining; I guess I can look with cool detachment as vendors seek to tempt me to update my library with with MQA or Hi-Res versions, confident/resigned to the fact that I will probably prefer the Redbook.

 

?

 

Mani.

[manisandher]

14 hours ago, Ryan Berry said:

It's always interesting how people hear different things when listening to the same music...

 

Thanks for sharing your thoughts Ryan.

 

14 hours ago, Ryan Berry said:

... I will also say that what I heard in these files is not nearly as distinct as other independent listening tests we've done with the formats.  I'm not sure if it's because of the conversion to .wav (I'm not in touch with the digital audio particulars as much as the engineers here), but I think any of these comparisons may need to be taken with a grain of salt...

 

The original hires FLAC file and the captured hires WAV file null exactly:

 

We can therefore assume that the capturing chain is essentially perfect.

 

And there is no reason to doubt that the Roon MQA decoder is doing its job correctly:

 

I apologise if the MQA capture sounds 'better than it has any right to' - but it has nothing to do with lack of resolution on the part of the capturing chain.

 

Mani.

[manisandher]

2 hours ago, sandyk said:

I wouldn't write off genuine high res just yet. Barry Diament's Soundkeeper Recordings will soon be releasing a new album in 16/44.1, 24/96 and a GENUINE 24 /192 resolution. It will be available in .aiff format which it was recorded in, as well as .wav files.

 

Alex, I would love to believe in genuine 24/192. But take a look at the spectrogram for 'Jason Vitelli - Confluence - 06 Say No More':

 

 

Which instrument has energy right up to 96 kHz (!)?

 

I did ask Barry (quite a few years ago now) what could be causing this, and he said he'd ask around... but he never got back to me.

 

Look, I'm not suggesting that this or any other hires is 'fake', in the sense that it's been deliberately derived from a lower resolution master. But the spectrograms do look odd, which suggests that something is off. (Oh, and one of those faint lines lies right on 88.2 kHz!)

 

Mani.

[manisandher]

Here's the spectrum/spectrograph for 2L's BEETHOVEN Op. 74 Harp Poco Adagio - Allegro in 24/384352.8:

 

 

Yes, there's definitely music >22.05 kHz, so a good case for hires perhaps. But this could be captured easily with an 88.2 sample rate. I mean, what on earth is the point of sampling at 384352.8 when there's nothing but noise above 44.1 kHz?

 

Oh, and that faint line lies exactly on 44.1 kHz (!).

 

Mani.

[manisandher]

2 minutes ago, PeterSt said:

IOW, I waited 5 minutes but I think you forgot something to post ...

 

2 minutes ago, PeterSt said:

So show us ... 

 

Should be up now.

 

Mani.

[manisandher]

1 minute ago, PeterSt said:

 

Yep. And exactly how I promised. Including no filtering.

So maybe the ADC's filtering is the bad boy ? Too bad it looks to be a music-frequency-correlated bad boy.

 

Maybe you didn't see this edit:

 

15 minutes ago, manisandher said:

Oh, and that faint line lies exactly on 44.1 kHz (!).

 

Mani.

[manisandher]

Just now, PeterSt said:

Looks odd to me after all. Something now is wrong. Including the rising slope beyond that point.

 

Exactly.

 

Looks like an original 88.2 'upsampled' to 352.8. But why???

 

Mani.

[manisandher]

23 minutes ago, Miska said:

The 44.1k line is probably word clock used to synchronize converters leaking at low level to the ADC.

 

Looks like the 88.2 sample-rate Nyquist line to me. Take a look at a close up of one of the peaks:

 

 

Surely that's clear imaging right there?

 

Mani.

[manisandher]

16 minutes ago, GUTB said:

Overall there is a lack of detail and a flatness which may be because of the track being re-recorded.

 

The track was NOT 're-recorded', it was captured digitally. The hires WAV capture nulls perfectly with the original hires FLAC. You will NOT hear a difference between the original hires FLAC and the hires WAV capture... unless your software player sounds different when playing FLAC vs. WAV (which it really shouldn't, if it's worth its salt).

 

Mani.

[manisandher]

2 minutes ago, Miska said:

I'd say it is nulls of the PCM decimation filter...

 

And this? (2L's Haydn String Quartet in D Op 76 No 5 - Finale - Presto - DXD):

 

 

What a total mess! And this from 2L!

 

Mani.

[manisandher]

3 minutes ago, Le Concombre Masqué said:

what's your point?

 

On 10/30/2018 at 7:04 AM, manisandher said:

I'm still totally done with all hires. It seems to me that if you can't get properly done hires from ECM/Qobuz, then you're not likely to get it from anywhere.

 

Let me know if you're still confused and I'll see if I can state it any more simply for you.

 

Mani.

[manisandher]

4 minutes ago, Ralf11 said:

listening impressions done right are the only std. for a consumer

 

listening impressions conflated with confirmation bias should be dismissed

 

There could have been no confirmation bias before I revealed the results... hence the purpose of this thread.

 

Mani.

[manisandher]

51 minutes ago, Miska said:

If you want something from 2L that has flat noise floor (modulator noise completely filtered out), take a newer recording that has been made using Merging Horus.

 

From 'apples-to-apples II':

 

On 10/21/2018 at 11:57 PM, Miska said:

One piano album I like is Jan Gunnar Hoff's Living, recorded by 2L. There's a sample track in original DXD and other formats here:

http://www.2l.no/hires/index.html

 

There's a nice video of the same track here:

 

 

Did this use the Horus? If so, I'm not impressed - there's as much noise at 176.4 kHz as there is actual signal between 3-6 kHz:

 

 

Why would this recording require anything more than redbook?

 

Mani.

[manisandher]

33 minutes ago, sandyk said:

Barry's microphones are -1dB at 40KHZ, and his recordings have genuine music content as high as 57KHZ.

 

Thanks Alex. I have absolutely no doubt. So where does the signal going all the way to 96 kHz come from? Something is amiss.

 

Mani.

[manisandher]

4 minutes ago, Miska said:

Well, that would need at least 88.2k sampling rate.

 

But there's virtually no signal above 22.05 kHz.

 

Mani.

[manisandher]

35 minutes ago, Miska said:

To me it looks like the content fades to background noise around 40 kHz. But you have very little scale on X-axis so it is hard to say.

 

More like 26 kHz. But the signal peaks at around -125dB at 22.05 kHz:

 

But just look at all that HF noise you'd be putting through your system! For what?

 

'JGH_take-111_raw-DXD' seems perhaps worthy of a higher-than-redbook sample rate:

 

 

Though the signal above 22.05 kHz remains vanishingly small. At least the HF noise is under control with the Horus.

 

Mani.

[manisandher]

2 hours ago, Le Concombre Masqué said:

... you seriously fed my suspicion that you might have a business agenda with MQA Ltd.

 

Your problem, not mine. Get over it.

 

Mani.

[manisandher]

Getting back to the track at hand, here's the spectrum/spectrogram of the first minute of Sample B (original hires), before the cymbals kick in:

 

Is it possible to record an instrument like this...

 

... and produce output that extends up to 48 kHz?

 

If the answer is 'yes', then all may be good with the track.

 

Mani.

[manisandher]

17 minutes ago, Miska said:

But what does this have to do with MQA?

 

It's in the title of the thread - 'MQA vs. hires'. I'm trying to establish whether the track I used in this thread is a 'legit' example of hires, after earlier comments such as:

 

On 10/28/2018 at 4:21 PM, PeterSt said:

 

Thank you for this.

What troubles me is that there is no decay in the highs (towards the higher frequency). This can't exist.

 

Maybe you forgot to show the end cymbals. Of course you could argue that the ride on the (ride-)cymbal shows ever so much more frequency, but ... 45KHz ? No way. And at this (soft ticking) level ? triple no.

The end cymbals oddly enough show about right to me.

https://www.musical-u.com/learn/percussion-frequencies-part-2-cymbals/

 

I did not emphasize anything with n dB and I always compare at this level. So the Greendale I showed - same thing. I never touch the level (in no 10 years of time) thus both are comparable. 

 

Also, I looked into this this morning because if I had to be correct on the how I came to my judgment on the Hires (sounds the worst) then first I had to be correct on it being fake, right ?

 

Peter

 

 

 

On 10/29/2018 at 3:28 AM, pkane2001 said:

By the way, I neglected to post the cepstrum plot for the comparison of MQA and Hires. It does point fairly strongly to a Hires file that was manipulated from a 48KHz original.

 

Please see the large peak at 24KHz and an additional +/-4KHz ringing around it (blue is original Hires, pink is MQA):

 

What's interesting is that after MQA processing, some of the ringing has been reduced (same plot, but MQA is now on top):

 

 

 

[Highlights mine.]

 

Mani.

[manisandher]

26 minutes ago, sapporo said:

Samples contaminated.

 

Where's your evidence for this?

[manisandher]

15 minutes ago, sapporo said:

Didn’t you write that one sample can possibly be uprezed hi resolution?

 

No.

[manisandher]

12 minutes ago, sapporo said:

To compare "apples to apples" it is necessary to be 100% sure that the "mother master mix" is the same.

 

Agreed.

 

Analysis (which anyone can do with the right tools) shows that the three samples in this thread are from the same master. For example:

 

On 10/26/2018 at 8:57 AM, esldude said:

I would agree they seem to be the same master having examined them some.  

 

I am '100% sure that the "mother master mix" is the same'.

 

Mani.

[manisandher]

1 hour ago, Rexp said:

Try this recording by same artist, has more life to it:

"The Astounding Eyes Of Rita" by Anouar Brahem http://tidal.com/album/14338143

 

 

From a purely technical POV, the two albums have very similar dynamics. Here's a comparison of the track used in this thread vs. one from the album you mentioned:

 

Opening Day:

 

The Astounding Eyes of Rita:

 

The LRA is a good representation of the music track dynamics, and they're identical in both cases. But of course, our ears might hear things differently.

 

Mani.

[manisandher]

17 minutes ago, austinpop said:

Or are you saying that only in the context of the 3 samples you've distributed?

 

Rajiv, with regards to the hires used in this thread, I'm happy with Miska's possible explanation for its extended HF content, namely:

 

On 11/1/2018 at 7:36 AM, Miska said:

Yes, likely [signal up to 96 kHz] is, with that kind of mic setup. Not the instrument itself after you have released the string, but the sound of nails hitting the strings can produce fairly sharp transients which you would see as horizontal lines in your vertical flowing spectrogram.

 

22 minutes ago, austinpop said:

Have you and Peter concluded that ALL hi-res content is flawed, and we should all just go back to Redbook and be happy?

 

I've not been able to draw any conclusion about whether all hires content is flawed or not. A lot of hires spectra do look odd though.

 

I've spent the last couple of days listening to hires vs. redbook I have from the same master, and can only say that I'm generally totally fine with the sound of the redbook. But I'm using PeterSt's software player and DAC, which have been specifically designed to optimise redbook (though not at the expense of hires).

 

Right now, I have no desire to pursue hires any further.

 

HTH.

 

Mani.

[manisandher]

1 hour ago, Arpiben said:

I would like to be sure of the origin of noise floor encountered in the sample A (24/96):

 

It may be useful to see the spectra from Samples A and B superimposed:

 

 

Up to ~16 Khz, the spectra look similar. Above ~22.4 kHz, Sample A (16/44.1 converted to 24/96) has a noise floor of -136dB, whereas Sample B (hires) has a signal of ~-116dB.

 

Does this help?

 

Mani.