MQA Listening Impressions

[manisandher]

2 minutes ago, PeterSt said:

Instead it is a good idea to investigate spectrographs.

 

 

Original 24/88.2:

 

16/44.1:

 

Mani.

[PeterSt]

No, we can't so much with that either.

[PeterSt]

Mani, what do you thing you actually see here ?

 

 

IOW, what is happening where or not etc. ?

 

 

[manisandher]

20 minutes ago, PeterSt said:

IOW, what is happening where or not etc. ?

 

These seem to match the FFTs I put up earlier.

 

The 24_88.2 falls rapidly after 40kHz because of the effective filter in HQPlayer - it's 128dB down at 45kHz. The MQA doesn't fall so much because of XXHighEnd's Arc Prediction filter, producing images above 48kHz (edit: 44.1kHz) - it's 108dB down at 45kHz.

 

And both spectra start rising at 45kHz because of the ADC's noise-shaping.

 

No?

 

Mani.

[matthias]

Guys,

what is the point?

The thread is about listening impressions.

 

Matt

 

 

[manisandher]

2 hours ago, PeterSt said:

Mani, what do you thing you actually see here ?

 

 

IOW, what is happening where or not etc. ?

 

Take 2...

 

In the 88.2 non-MQA, that's quite a lot of 'content' going up to and well beyond 40kHz, and it's pretty even across the whole track. How many mics have this sort of range? And how many instruments produce content to beyond 40kHz?

 

This then causes more imaging in the MQA file than you might otherwise expect, causing potential problems in the sound.

 

???

 

Mani.

[manisandher]

3 minutes ago, matthias said:

Guys,

what is the point?

 

 

3 minutes ago, manisandher said:

This then causes more imaging in the MQA file than you might otherwise expect, causing potential problems in the sound.

 

 

3 minutes ago, matthias said:

The thread is about listening impressions.

 

Yep, but if the 'reference' file is a dud, then there's a problem. It may also explain why you and a bunch of others prefer the 16/44.1 capture...

 

Mani.

[Miska]

23 minutes ago, manisandher said:

In the 88.2 non-MQA, that's quite a lot of 'content' going up to and well beyond 40kHz, and it's pretty even across the whole track. How many mics have this sort of range? And how many instruments produce content to beyond 40kHz?

 

There's quite a lot of content that reaches around 40 - 50 kHz filling the entire band of 96 kHz sampling rate. There is absolutely no point in "hires MQA" if there's no content in the 22.05 - 44.1k band.

 

There are three frequently used wide-band omni microphones from DPA (for example 2L uses those):

http://www.dpamicrophones.com/microphones/ddicate/4006a-omnidirectional-microphone

http://www.dpamicrophones.com/microphones/ddicate/4007a-omnidirectional-microphone-12mm

http://www.dpamicrophones.com/microphones/ddicate/4003-omnidirectional-microphone-130-v

Of which the 4007a is quite staggering by going pretty much flat to 40 kHz while of course not stopping there yet. These were originally branded Brüel & Kjær, but later split to a separate company.

 

And then Sennheiser has equivalent:

https://fi-fi.sennheiser.com/recording-condenser-microphone-onmi-directional-guitar-acoustic-bass-brass-mkh-8020

 

Widest band is Sanken CO-100K:

http://www.sanken-mic.com/en/product/product.cfm/3.1000400

 

[Miska]

3 hours ago, PeterSt said:

Mani, what do you thing you actually see here ?

 

 

IOW, what is happening where or not etc. ?

 

You can see that actually the MQA content begins to roll off earlier, content is getting visibly fainter already at 25 kHz, while the there's lot of images still beyond Nyquist of the source (44.1 kHz). So a very slow slope.

 

Original looks like a normal hires recording that actually has some HF content, so not fake.

 

[fas42]

I asked a short whole ago about whether a "normal" musical track with content above 20k could be pointed to, where there was anything meaningful or audible within that was strictly musically related - and got no answer. Which pretty well says where this 'discussion' is at ...

 

In context, 2L recordings had >20k stuff where things in the recording space were accidentally knocked, or right in the middle of a musical climax, buried under a cacophany of general sound.

[Miska]

3 minutes ago, fas42 said:

I asked a short whole ago about whether a "normal" musical track with content above 20k could be pointed to, where there was anything meaningful or audible within that was strictly musically related - and got no answer. Which pretty well says where this 'discussion' is at ...

 

I even gave you link to one? How many you want? You can also check out the graphs published by Hifi-News every month for bunch of hires recordings. Getting responses going up to 50 kHz is not at all unusual for modern micrphones.

 

Another one I've been using for MQA vs non-MQA comparisons is this blues track:

 

[Samuel T Cogley]

Can anyone recommend a Tidal/MQA title that isn't (too) peak limited that I can use with my new iDSD Nano BL?

 

 

[fas42]

Just now, Miska said:

 

I even gave you link to one? How many you want? You can also check out the graphs published by Hifi-News every month for bunch of hires recordings. Getting responses going up to 50 kHz is not at all unusual for modern micrphones.

 

 

Sorry if I missed the post, . I look at precisely what the context of the sound is, and how loud the rest of the spectrum is, at the actual moment - and whenever I do that the hi res content is so far down compared to all else, or is incidental; I want a piece where the hi res is front and central, and part of the musical message.

[Miska]

5 minutes ago, fas42 said:

Sorry if I missed the post, . I look at precisely what the context of the sound is, and how loud the rest of the spectrum is, at the actual moment - and whenever I do that the hi res content is so far down compared to all else, or is incidental; I want a piece where the hi res is front and central, and part of the musical message.

 

Anything that has close mic'ed cymbal kit with good mics pretty much fills the bill.

 

In order not to have those time domain effects, the sampling rate needs to be high enough that all the musical content harmonics have disappeared in noise floor before the decimation filter's transition band begins. Sometimes 88.2/96 kHz sampling rate is enough for that, sometimes it is not. In many cases 176.4/192 kHz is enough though. And you pretty much need that Sanken microphone to get close to DXD limits.

 

[Miska]

I've recorded glockenspiel:

 

And maracas:

 

Using a recording mic that is documented to be flat to 20 kHz. It goes quite a bit above that before rolling off. But the glockenspiel was enough to nicely demonstrate the filter ringing effect when converted to 44.1 kHz sampling rate because it has nice fast attack. Maracas on the other hand will put lossy decoders to shame with it's intense noise-like signal. I have bunch of other stuff too.

 

These files would work quite OK at 96 kHz sampling rate, but not at 44.1 kHz.

 

Now I'm planning to get pair of those Sennheiser mics for further testing.

 

[Miska]

Thank you @manisandher for the DSD versions too! Listened through direct upsampling to DSD256 it is clearer! Same listening conclusions as before though.

 

Here are my plots generated from the DSD files.

 

Original hires:

 

MQA:

 

[mansr]

1 hour ago, Miska said:

Maracas on the other hand will put lossy decoders to shame with it's intense noise-like signal.

The "castanets" test file was also great for tormenting lossy encoders.

[mansr]

58 minutes ago, Miska said:

Here are my plots generated from the DSD files.

 

Original hires:

 

MQA:

High-frequency images clearly visible.

[PeterSt]

3 hours ago, mansr said:

4 hours ago, Miska said:

Here are my plots generated from the DSD files.

 

Original hires:

 

MQA:

Read more  

High-frequency images clearly visible.

 

Yes, but in "unfolded" fashion. Literally !

Later I will try to point it out.

[matthias]

14 hours ago, matthias said:

 

Mani,

some honest comments from another forum:

 

http://www.lejonklou.com/forum/viewtopic.php?f=2&t=3970

 

Matt

 

Mani,

now updated with a new listening impression.

 

Matt

[manisandher]

4 hours ago, PeterSt said:

Yes, but in "unfolded" fashion. Literally !

 

Hey Peter, in the MQA capture I definitely had the MQA decoder switched on in XXHighEnd to unfold the MQA file from 44.1 to 88.2.

 

(I've just checked the non-decoded MQA capture that I also took just as a test, but obviously never posted. It looks almost identical to the 16/44.1 capture, which is exactly as we would expect.)

 

You must have something else on your mind...

 

Mani.

[manisandher]

17 hours ago, PeterSt said:

But then the source (I mean files) must be unambiguously be trusted. Right ?

 

It looks like the original 24/88.2 file was created for Erato by 'Soundways' in Paris, http://soundways.fr/en/.

 

Looking at the sort of material Soundways work on, I would say the 24/88.2 original file is legit - I really can't imagine them manipulating it to make it look 'more hires' than it actually is. And I can't see HDTracks going to these lengths.

 

My conclusion: a well-recorded track with lots of real HF content, and perfect to test how MQA sounds.

 

Mani.

[matthias]

34 minutes ago, manisandher said:

My conclusion: a well-recorded track with lots of real HF content, and perfect to test how MQA sounds.

 

Agree.

 

Matt

[Miska]

13 hours ago, mansr said:

The "castanets" test file was also great for tormenting lossy encoders.

 

Yes, that's what I actually wanted to say (encoders). Thanks for correcting!

 

[manisandher]

On 11/30/2017 at 7:01 PM, PeterSt said:

But then the source (I mean files) must be unambiguously be trusted. Right ?

 

8 hours ago, manisandher said:

My conclusion: a well-recorded track with lots of real HF content, and perfect to test how MQA sounds.

 

Nah... I think Peter was correct.

 

Take a look at the spectrum for another track on the album:

 

 

There is clearly something weird going on at 42kHz.

 

It's harder to see, but there's something similar in the track I used for the captures:

 

 

It looks to me as though on the original 24/88.2 file, all the content above 42kHz is fake.

 

Thoughts?

 

Mani.